Documente Academic
Documente Profesional
Documente Cultură
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry.
http://www.jstor.org
Mark Liberman,
Alan Prince
0. Prospectus
In this essay a new theory of stress and linguistic rhythmwill be elaborated,based on
the proposals of Liberman(1975).' It will be argued that certain features of prosodic
systems like that of English, in particularthe phenomenon of "stress subordination",
are not to be referredprimarilyto the properties of individualsegments (or syllables),
but rather reflect a hierarchical rhythmic structuring that organizes the syllables,
words, and syntactic phrases of a sentence. The characterof this structuring,properly
understood, will give fresh insight into phenomena that have been apprehendedin
terms of the phonological cycle, the stress-subordinationconvention, the theory of
disjunctive ordering,and the use of crucial variables in phonologicalrules.
Our theory will employ two basic ideas about the representationof traditional
prosodic concepts: first, we represent the notion relative prominence in terms of a
relation defined on constituent structure;and second, we represent certain aspects of
the notion linguistic rhythm in terms of the alignment of linguistic material with a
"metricalgrid".
The perceived "stressing" of an utterance, we think, reflects the combined
influence of a constituent-structurepattern and its grid alignment. This pattern-grid
combination is reminiscent of the traditionalpicture of verse scansion, so that the
theory as a whole deserves the name "metrical". We will also use the expression
"'metricaltheory" as a convenient term for that portion of the theory which deals with
the assignment of relative prominence in terms of a relation defined on constituent
structure.
Section 1 will apply the metricaltheory of stress-patternassignmentto the system
of English phrasalstress, arguingthis theory's value in rationalizingotherwise arbitrary
characteristicsof stress features and stress rules. Section 2 will extend this treatmentto
the domain of English word stress, adopting a somewhat traditional view of the
assignment of the feature [+stress], but explaining the generation of word-level
* We would like to thank J. B. Grimshaw, M. Halle, S. J. Keyser, R. P. V. Kiparsky, D. L. Nanni, E.
0. Selkirk, and J.-R. Vergnaud for much valuable discussion of the materials presented in this article.
1 Aspects of these proposals, or ideas similar in spirit, are to be found in Fischer-J0rgensen (1948),
Rischel (1964; 1972). This previous work will be discussed at the end of section 3.
250
MARK
LIBERMAN
AND
ALAN
PRINCE
For the most part, this article will advance its case in terms of a reanalysis of old
observations, rather than a flourish of new ones. The body of descriptive data we
propose to reinterprethas been common (differences of detail aside) to the linguistic
traditionthat includes the AmericanStructuralistsas well as GenerativePhonology; it
includes the distribution of stressed and unstressed syllables in English words, the
location of main word stress, the differential treatment of lexical and supralexical
constituents, the preservationof relative prominenceunderembedding,and so forth.
Over the years, some fundamentalcharacteristics of this description have been
called into question on one basis or another. For example, it has been proposed (most
notably by Bolinger)that it is incorrectto extend the notion "stress" beyond the word
level, "sentence stress" being merely a matter of pitch-accent placement. This same
point of view holds that word stress itself is no more than a guide to the word-level
location of those (phrasal)pitch-accentsthat the speaker chooses to impose.
Although it is not our purpose here to argue the point, we feel that the
structuralists(and their generative heirs) were closer to the truth. English is a stress
language, not a tone or pitch-accent language; English stress patterns, within and
among words, have phonetic reality as rhythmicpatterns entirely independentof their
role in orchestratingthe placement of intonationcontours.
This much should not be difficult to establish experimentally.One promisingline
of inquiry relies on the fact that it is possible to mimic an arbitraryEnglish utterance
while substituting reiteration of a single syllable (e.g. ma) for each syllable of the
original. Such "reiterant speech" shows stable durationalpatterns, which depend on
the stress patternand constituent structureof the utterance,2just as durationalpatterns
in natural speech do. It has been shown3that listeners are able to extract stress and
constituent-structureinformationfrom reiterantspeech, and that (underthe conditions
of the cited experiment)durationis the dominantcue in both cases.
In perception experiments, the use of reiterant speech guarantees that stresspattern perceptions cannot be derived from the hearer's knowledge of the words
involved; control of Fo, possible by using analysis-resynthesis techniques, can be used
to separate out the role of "pitch accent" (a role that is perhaps less central to stresspattern perception than is generally believed). In analysis of production data, the
2 Cf. Liberman and Streeter (1976).
1 By Nakatani and Schaffer (1976).
ON
STRESS
AND
LINGUISTIC
RHYTHM
These two observations (suitably expanded) lie at the root of the traditionalaccount,
within generative phonology, of English phrasal stress. We propose to substitute a
theory that embodies the same observations in a rather different way; in order to
understandthe natureand value of the differences, we begin by describinghow English
phrasal stress has been treated within a familiargenerativetheory of phonology.
This theory employs an n-ary segmental stress feature, that is, a phonological
feature, otherwise like other distinctive features of segments, which is in principle
capable of assuming indefinitelymany values. Its range is usually limitedto five values
(four levels of stress, from 1 to 4 in decreasing strength,plus [Ostress]) more or less as
a matterof convenience.
The idea of a stress feature with somethinglike three to five values should not be
laid, either as credit or debit, to the account of generative phonology. Steele (1775)
employed the symbols A, .-., and: to markthree levels of stress within the barlinesof a
quasi-musicalnotation (which also markedpitch contours and seven degrees of length);
the use of symbols such as ', , ^, , I, and I to specify three to five levels has long been
I By "phrasal" stress we mean stress above the level of the word, including the stressing of lexical
compounds as well as that of truly "phrasal" categories.
252
MARK
LIBERMAN
AND
ALAN
PRINCE
common; and the numbers 1-4 were used by the American Structuralistswhose work
formed the foundationon which Chomsky and Halle built.
An essentially novel contributionof generativephonologicaltheory was the idea of
cyclic reassignmentof [1 stress], coupled with the convention of stress subordination,
which follows the reassignmentof [1 stress] in a given domain, and demotes all other
stresses within that domainby one level. Let us observe these principlesin operationin
a simple case. In the first line of (1), each word is assumed to have been providedwith
the feature [1 stress], on the appropriatesegment, by the word-level stress rules (to
which we will returnin section 2). The principleof cyclic applicationof rules tells us to
begin with the innermostset of brackets, which in this case defines the constituentdewcovered.
Inner Cycle-CSR
OuterCycle-NSR
Q c]
[1 stress]
Conditions: Q contains no [1 stress]
C = NP, VP, AP, S
ON
b. CSR
V
[1 stress]
STRESS
[I stress]/
AND
LINGUISTIC
RHYTHM
Q (##P) c]
(3) a. [#dew##covered#]
I
-Q##
1
~~~~~~~~~~I
I
b. [#dew##covered##lawn#]
VI
- Q]
On the noun phrase cycle (representedin (3b)), the NSR will be selected; Q will cover
the sequence /n#/, and the rule will reassign [1 stress] to the vowel of lawn.
Let us now examine a few more complicatedcases. The predictionof the CSR, as
formulated in (2b), is that any compound with a uniformly left-branchingstructure,
that is, in which each new cyclic domain is augmentedby precisely one word at the
rightmargin,should retainprominenceon its leftmost member;however, if at any stage
of the compoundingprocess the righthandelement is itself a compoundform, then this
righthandmember will assume the primarystress.
For an example of the first sort, consider the compound law degree requirement
changes.
1
(4) a. [#lw
T#degree#]
IQ##
1
]
2
Output
On the first cycle, representedin (4a), the bracketin the rule analyzes the bracket
in the target string; the variable P analyzes the single word degree, being requiredto
254
MARK
AND
LIBERMAN
ALAN
PRINCE
(4) b. [#Iaw##degree#requirement#]
Q ##
1
Output
At the next higher cyclic level, represented in (4b), P analyzes the word
requirement; Q will take the maximal string containing no [1 stress], which is
/##degree/; and [1 stress] will again be reassigned to law, resulting in the patternlaw
degree requirement.
3
(4) c. [#w##degree##requirement##changesI
parallel. In each case, [1 stress] will be assigned to the [1 stress] vowel of the word on
the left, yielding a 1 2 1 2 pattern as input to the applicationof the rule at the upper
level.
1
(5) [#law##degree##lnguag
Q
2
##requirement#]
##
P
3
As usual, P will analyze a single word, in this case the word requirement. But Q is
constrainedto contain no [1 stress], and thereforeits maximaldomainwill be the string
/hgwij/; the rule will reassign [1 stress] to the [1 stress] vowel of language. This result
is in accord with the facts-compounds of this general type, such as labor union
finance committee, fly-rod trout-fishing, etc., typically have their main stress on the
ON
STRESS
AND
LINGUISTIC
RHYTHM
In the "traditional" theory we have been discussing, the patterns of stress numbers
producedby applyingthe stress rules to syntacticallymotivated surface structureshave
been assumed to be subject to various sorts of correction.
In some cases, the desired result has been obtained by assuming that surface
syntactic structuresare modifiedby certain "readjustmentrules" before being submitted to the phonological component, which will then produce the correct stress pattern
in the normal course of its operation. For example, a left-branchingstructuresuch as
5
[[[[John's mother's] brother's] dog's] house], which would produce the output John's
4
of all nonprimarystresses within the confines of a given word; thus Tennessee butAral
1
the same phrase, e.g. thirteen men or Tennessee Ernie. This has been attributedto a
"rhythm rule" operating to alleviate clash of stresses and produce a more nearly
alternating pattern. Except for the attempt in Kiparsky (1966), this rule has been
discussed only informally.5
In section 1.3, we will present a different treatment of English phrasal stress
assignment. This treatmentwill employ the first of the concepts we hope to motivate in
this article, the idea of representingrelative prominence in terms of a binary relation
I Goldsmith (1976)
attempts to explain such cases as misinterpretation of certain properties of tonecontour assignment rules. The grounds for our rejection of this view will become clear in section 3.
256
MARK
LIBERMAN
AND
ALAN
PRINCE
b.
w s
I II
red cows
c.
d.
/A\
s
w s
John left
s
lw
keel haul
stress shift
(7)
w
dew-covered lawn
6 We use the symbol "R" as a convenient label for the root of the tree, which will of course be neither s
nor w, since it is not in a syntagmatic relation with any other node.
ON
STRESS
AND
LINGUISTIC
RHYTHM
In other words, stress subordinationis built into the basic nature of this mode of
representation,and does not requireany special convention.
Another effect of assigningphrasalstress patternsin terms of structuralratherthan
segmental features is that the stress-assignmentrules now define strictly local properties of their input. It is not necessary to hunt through a segmental string for "the
rightmost [1 stress] vowel," or "the rightmost [1 stress] vowel to the left of the
rightmost ## boundary." Rather, we simply define a relation on each pair of sister
nodes in the syntactic structure,the outputdependingon certain local propertiesof that
structure. To be specific, this theory's version of the NSR and CSR can be stated as
follows:
(8) In a configuration[cA Bc]:
a. NSR: If C is a phrasalcategory, B is strong.
b. CSR: If C is a lexical category, B is strong iff it branches.
Because of the relationalnature of the categories "strong" and "weak", making
one element strong necessarily makes its sister weak, and vice-versa. Given that some
pattern of relative prominence must be assigned by every applicationof the rule, the
effect of the biconditionalin the above statementof the CSR is thus to guaranteethat if
B is nonbranchingit will be weak (and A, correspondingly,strong).
Now, let us reconsider the derivation of law degree requirement changes:
(9) a.
2
3
law egree requirementchanges
b.
R
s
yw
w
yw
law degree requirementchanges
The nodes labelled 1, 2, and 3 in (9a) requirethe s/w relationto be defined on their
daughters. In each case the parent node is the lexical category noun, so the CSR will
be employed throughout.In every case the right daughteris nonbranching(consists of
a single word), so a trochaic patternwill be defined in every case, as shown in (9b).
258
MARK
AND
LIBERMAN
ALAN
PRINCE
(10) a.
1
2
R
w
( 1) [#law##degree##requirement##changes#]
-
I
I##I \1,~
P
]
2
Analysis
Output
We will have more to say later about the phonological cycle-for now we wish
simply to observe that it is a formal necessity in the segmental treatment of stress
patterns, given the use of the stress subordinationconvention as a way to representthe
preservation of relative prominence under embedding. The theory we are proposing
encodes relative prominence directly, as a local feature of constituent structure, and
therefore the formalismitself does not force a cyclic procedureof rule application.
Let us now pause briefly to consider what the labellings we have been imposing
really mean, in terms of patterns of relative stress. The "most prominent" terminal
ON
STRESS
AND
LINGUISTIC
RHYTHM
element of a given constituent must surely lie somewhere in its "more prominent"
daughter,an argumentthat can be applied recursively until the most-stressed terminal
element is reached. Thus for any given constituent, the location of its main stress (as
far down as the lowest level of arborization)can clearly be found by following a path
from the root that intersects no nodes labelled w. We will call the most prominent
terminal element of a given constituent, its "main stress", by the name designated
terminal element.
It is less clear how such trees should be considered to define relative prominence
among their non-main-stressedterminal elements. If we wished to mimic closely the
numerologyof previous theories, we could make use of the following definition:
(12) If a terminalnode t is labelledw, its stress numberis equal to the numberof
nodes that dominate it, plus one. If a terminalnode t is labelled s, its stress
number is equal to the number of nodes that dominate the lowest w
dominatingt, plus one.
To illustrate the method of mimicry, we offer the following sufficiently complicated compound:
R
(13)
II\\
s
So far, we have examined only cases in which the cyclic reassignmentof [1 stress], by
the rules in (7), and the definitionof the strong/weakrelationon sister constituents, by
260
MARK
LIBERMAN
AND
ALAN
PRINCE
the rules in (8), are weakly equivalent (i.e. the two approaches establish the same
patternof relative prominence).We will argue, in section 1.5, that the relationaltheory
rationalizesa numberof otherwise arbitrarypropertiesof such patternsand of the rules
that define them, and is thereforeto be preferredon the grounds of greaterexplanatory
adequacy. However, there is at least one class of cases in which the two theories differ
descriptively. Unfortunately, the crucial examples cannot normally be constructed in
English, but it is worthwhile to note the point at issue, and to examine a case
(admittedlyratherunusual)where the descriptive difference matters.
This descriptive difference will arise when the CSR applies to a constituentwhose
righthandmember, itself complex, has main stress on its rightmostword. This situation
is representedschematicallyin (14), first as it appearsto the relationalrule, and then as
it appearsto the numerologicalrule.
(14) a.
b.- . . # N##NM#]
##P
Q
1
Analysis
Output
In such a case, the relationalCSR will define greatest prominenceon the rightmost
word (representedas "M"), while the numerologicalCSR will reassign [1 stress] to the
(rightmost)[1 stress] of the lefthandsubconstituent(representedas "Y" in (14a) and as
the string of dots in (14b)).
To see why this will be the result, compare the formulationof the relationalCSR,
in (8), with that of the numerologicalCSR, in (2b).
The relational CSR says that the righthandsubconstituent will be strong if and
only if it branches, that is, if and only if it consists of more than one word. Its operation
is not dependent on the way the strong/weakrelation is defined at any other level of
constituent structure, and so the node labelled "Z" in (14a) will become the strong
memberof the pair [YZ], whether Z is internally[ws] or [sw].
The numerologicalCSR, on the other hand, always analyzes the rightmostword in
its domain as the variableterm P, and then scans the remainingstringfor the rightmost
occurrence of [1 stress]. If the word analyzed by P contains the [1 stress] vowel of its
constituent, then that constituent as a whole has lost its chance to have [1 stress]
reassigned within it, regardlessof the numberof words it contains.
ON
STRESS
AND
LINGUISTIC
RHYTHM
Because of the way English compounds are constructed, real cases meeting the
above specifications are ratherhard to come by. In the normal course of events, they
are impossible, since a collocation of words normally has main stress on its final
member only if it is a phrasalconstituent, and compounds may be built up only out of
2
that takes a rising stress pattern, such as Madison Avenue, or an adjective + noun
grouping that has been lexicalized with its original stressing, while still retainingan
internal ## boundary (unlike e.g. blueprint, which both takes initial stress and also
functions as a single word from the point of view of the rules we have been
considering).
We have located an actual occurrenceof a fairly clear case of the latter(lexicalized
adjective + noun) type, namely the expression motor unit neural control, meaning 'the
neural control of motor units', and thus parsed as [motor unit] [neural control]. A
motor unit consists of a motoneurontogether with the muscle fibers it innervates;the
word motor unit is stressed as a compound. In the milieu in which this example was
found, "neural control" is such a central topic that the phrase has apparentlybeen
lexicalized. As evidence for this analysis of the situation, we observe that compounds
such as motor unit control, 'the control of motor units', are common and unexceptionable, while a collocation involving a clearly phrasalrighthandmember, e.g. *motorunit
higher-levelcontrol for 'the higherlevel control of motor units', seems ill-formed.Thus
motor unit neural control must be a compound form, and is therefore a plausible
example of the kind of case we need. In our opinion, and that of others we have polled,
the main stress of this phrase, used for example in the frameOurlab is interestedin the
problem of motor unit neural control, would tend to fall on the word control, in
262
MARK
LIBERMAN
AND
ALAN
PRINCE
segment elsewhere in the string, so that the monosyllableJohn, for example, could not
conceivably stand by itself as a phonological representation. The values of other
phonologicalfeatures are generallydefined paradigmatically,that is, strictly in terms of
an opposition with a differentfeature specificationthat could have occurredin the same
segmental position.
(iii) The distinction among various levels of stress has little or no local phonetic
import-the feature [1 stress] does not imply any specific articulatory or acoustic
property of the segment that bears it, any more than the notion "downbeat"
necessarily defines any intrinsicpropertyof certain notes in a piece of music, by which
they could be sorted without reference to some fairly extensive context. This is not to
say that distinctions in degree of stress have no phonetic effect, but simply that their
effect must be defined in terms of a pattern (of pitch and time relations)that generally
extends over a stretch of utterancemuch largerthan the roughlysyllable-sizeddomains
in which the phonetic implementationof other phonologicalfeatures is to be found.
(iv) The preservation of relative prominence under embedding has provided the
clearest evidence for cyclic application of phonological rules. Most of the other
evidence for the phonological cycle has come from other characteristicsof stress or
stress-relatedphenomena.7
(v) As a result of the stress-subordinationconvention, the effect of stress rules is
to cause a widespread pattern of change, rather than simply to change the feature
specification of a single segment, as is the case with most nonprosodic phonological
rules.
(vi) Stress-assignmentrules have typically made use of crucial variables, that is, a
variable term in the structural description of the rule that permits its locus of
application to be indefinitely far away from some other term necessary to define its
environment. Thus the [1 stress] vowel mentionedby the NSR may be any numberof
segments away from the constituent boundary required in the rule's environment.
Other phonological rules generally require either no variables, or variables of a
different type (called "segmental variables", e.g. C0), whose effective range is limited
to adjacent syllables.
(vii) Stress rules often make use of the convention of "disjunctive ordering". A
7 But see Kean (1974), Mascaro (1976).
ON
STRESS
AND
LINGUISTIC
RHYTHM
term in the environmentof the rule is placed in parentheses;this stands for an ordered
pair of rules, the first one includingthe parentheticalterm and the second one leaving it
out. This sequence of subrulesis ordereddisjunctively, meaningthat if the first subrule
applies, then the second subrule is skipped. Many familiar stress rules (e.g. that of
Latin) have disjunctively applicablecases, in which the environmentof one case is a
substringof the environmentof another, but such a notation is not motivatedfor rules
assigning other phonological features.8
These seven distinctive characteristicsof the segmentaltreatmentof stress patternsdo
not in themselves constitute any argumentagainst that treatment. However, we feel
that a relationaltreatmentof stress patterns,along the lines we are presentinghere, has
the advantage that it rationalizes these special characteristics of stress features and
rules, in the sense that they follow directly from the way the phenomenon is
represented,ratherthan being arbitrarytypological observations.
(i) In a relational theory, the stress feature is no longer n-ary, but (in a sense)
binary. It is, however, a relational feature of constituent structure rather than an
intrinsicfeature of phonologicalsegments; the rest of its special propertiesfollow from
this.
(ii) A relation of the kind we are proposing must by its nature be defined
syntagmatically,since constituentstructureis an essentially syntagmaticnotion. Thus a
relationaltheory has no way to representa secondary or tertiarystress standingalone.
(iii) Since a relationaltheory defines relative prominence as a feature of constituent structureratherthan of phonological segments, it is quite naturalthat its phonetic
implementationshould be in terms of largerpatterns, ratherthan in terms of some more
localized articulatorygesture or sound quality.
(iv) The way relative prominenceis representedin a relationaltheory guarantees
that it will be preserved underembedding,so that cyclic rule applicationis not required
to account for this phenomenon.
(v) In a relational theory, the phonological effect of the rules defining relative
prominence is a local one (the definition of the strong/weakrelation on sister nodes).
No counterpartto the stress-subordinationconvention is required.
(vi) In a relational theory, the rules defining relative prominence are locally
conditioned. In the cases we have considered so far, the domain is a pair of sister
constituents, and the conditioning factors are the category of the parent and the
terminality(branchingor nonbranching)of the righthandsister. No variableis required,
since we are annotatingthe nodes of a tree structureratherthan locating segments in
order to adjust their feature specification.
(vii) The phenomenawe have discussed up to this point do not provide any useful
case of disjunctive ordering, so we will postpone discussion of this issue to section 2,
where a numberof relevant cases will be described.
8 See
264
MARK
LIBERMAN
AND
ALAN
PRINCE
In the treatmentof stress-patternassignment above the word level, there are thus
significantadvantagesto the representationof relative stress as a structuralratherthan
a segmental feature. However, for this representationto be adequate for the needs of
phonological theory, it is necessary to show that it can be extended in a useful way to
the treatmentof relative stress within words. In the next section, we attempt to show
that such an extension is not only possible, but desirable.
2. Words
Hierarchicalstress subordinationis as characteristicof words as it is of phrases and
1
compounds. The perceived array of prominence in words like execute and cognate
1
closely resembles that of compounds like labor union, dog days; phrases like red cow,
2
globed peonie are similar to such words as pontoon, arrange; the pattern of union
1
In terms of the theory being explored here, this can only mean that words have an
internal metrical structure in which syllables and groups of syllables are weighed
againsteach other. For words, as for phrases, the patternof subordinationsis known to
be essentially lawful, and we must expect to find a rule to distributenode labels below
the phrasallevel, just as we found a metricalversion of the NSR and CSR.
However, in dealing with words, as opposed to phrases, we cannot appeal to a
syntax of syllables that would design the trees for us, independent of prosodic
considerations. We must thereforediscover the relevant principles of construction.
2.1. Word-Trees
Consider first the simple situation in which a stressed syllable is weighed against
unstressed syllables. It accords quite directly with the intuition behind metrical
comparisonto regardthe stressed syllable as strong, its unstressed compeers as weak.
This, taken with the restrictionto binary branching,dictates tree shape and labelling
for words like labor, caprice, Pamela:
(l5){s w
labor
s
Ii.
capnce
s
s w ,w
pamela
The (+,-) marks indicate the value of the segmental feature [? stress] for the
vowels they are written under. Although in the examples cited s dominates only (+)
and w only (-), this perfect correlation cannot be maintained in general, since a
ON STRESS
AND LINGUISTIC
RHYTHM
[+stress] vowel may well be metrically weak, as words like gymnast, raccoon show:
(16) a.
b.
ws
gymnast
+
raccoon
0 1
and wi- shows itself when we compare modest with gymnast or balloon with raccoon.
(17) a.
s w
ws
balloon
modest
Metrically, modest and gymnast, balloon and raccoon, can only be identical, because
the membersof each pair have identicalpatternsof relative prominence.Examples like
these show that the familiarsegmental (or syllabic) distinction marked by the feature
[? stress] must be maintainedwithin metricaltheory.
We hypothesize, then, that the correlation between (s,w) and L(+,-) stress] is
given by the following implication:
(18) If a vowel is s, then it is [+stress].
By contraposition,(18) tells us that if a vowel is [-stress], it must be w. Principle(18)
will be regardedas a well-formednesscondition on metrical structures, functioningto
disallow the output configuration(19):
(19)
*I
V
[- stress]
Principle (18) may be paraphrasedby saying that only a stressed syllable may be the
strong element of a metricalfoot.
Observe that the condition (18) gives only one parsingfor Pamela, eliminatingthe
logically possible (20b,c):
(20) a.
b.
c.
Sw
s
ww
pamela
+
- -
pmelei
+
ws
*pamel a
+
_-
266
MARK
(21) a.
AND
LIBERMAN
ALAN
PRINCE
b.
I I .1
II .1
acacia
acacia
Our principles determine the labels, but allow both rhythmic divisions equally.
Although it is possible that such multiplicitycould exist and function, we submit that
only (21a) is correct, and that English metricalstructureis well-defined, in accord with
the following description:
(22) a. Every sequence of syllables
+-,
+--,
+---,
/AA
b. The syllable-dominatingtrees of provision (a) are organized into a rightbranchingtree whose root is associated with the syntactic node immediately dominatingthe entire word. The arrangementwill look like this:
N
#AL
.. .A
A#
(This description will be modified slightly in 2.4 below, when certain more
complex cases are considered.)
Imposing these patterns on acacia decides in favor of (21a), [[wa][fcacia]];similar
results are guaranteedfor the parsingof America:
RHYTHM
b.
(23) a.
AA
w s ww
america
ww
amenca
reconcili ation
?
+_
The next stop is -cil-, and here again a trochee will be called for; but further
arborization is also possible, and the two trochees will be joined into a higher-level
unit:
(25)
? Ws w
reconcili ation
+
+-
sws
wV
reconcili ation
+
+_?
Observe that in a word like execute, the first stop-at the stressed syllable -ute--will
not result in the creation of metrical structure; the next stop will arborize the (+,-)
sequence exec- and join it to the remaining -ute, generating the structure (27):
(27)
sw
I
execute
+
-?+
268
MARK
LIBERMAN
AND
ALAN
PRINCE
At this point it is appropriateto ask what principle, if any, governs the labellingof
"higher-level" nodes. This is equivalent to asking what law or laws of rankinghold
among stressed syllables in English words. S. Schane ((1972), reportedin Halle (1973)),
has found a rule that predicts the location of main stress for a central portion of the
vocabulary:primarystress falls on the first stressed syllable from the end of the word,
leaving out the final syllable. Schane's insight translates into metrical theory as the
following simple rule:
(28) Word Rule
(N2 branches)
pS
Applying the Word Rule to the incompletely labelled examples (26) and (27), we
achieve the following representations:
b.
(30) a.
W
S\
wN
SS
ww
/ \ / /\
sw s,vs w
euxecute
(1
2)
reconcil i ation
(2
3 1
(31)
W
s
I
w
I
S
s
I
.1
ex ec ut ion
There is one respect in which the Word Rule appears to transcend Schane's
original goal: the metrical formulationprovides the key to labelling every node in the
ON
STRESS
AND
LINGUISTIC
RHYTHM
tree, where Schane's rule aims only at specifying the location of main stress. Indeed, in
the metrical theory as presently constituted, the rule could hardly function otherwise,
since finding the main-stressed syllable involves labelling (as s) every metrical
constituent it belongs to, and thus will generally mean labellingthe entire tree. It might
therefore be expected that details of the relationshipbetween nonprimarystresses in
long words would shed direct empirical light on the geometry of word trees, in
particularon the higher constituent structure, confirmingor disconfirming,for example, the choice of right-branchingin (22b). However, this relationship is rather
unsteady, as noted by Kenyon and Knott (1953, xxv), and is heavily influenced by
rhythmic factors. When we turn to consider problems of rhythm and relative prominence in section 3 below, we will argue for an interpretationof the metrical strength
relation that does not strictly imitate the effects of the NSRICSR algorithm. The
Relative Prominence Projection Rule (104) will, in itself, flatten out the distinctions
between secondary stresses in many configurations-e.g. in words like Ticonderoga,
prognostication,
touched on below in section 3.5. The function of the Word Rule, then, like that of
Schane's rule, is to identify primaryword stress; it accomplishesthis by examiningand
annotating a hierarchicalgrouping of syllables that is projected from the location of
stressed and stressless elements.
The astute reader will have noticed that the Word Rule (31) is identical to the
Compound Rule (8b). This result seems a sure sign that pursuing the arboreal
imperativebeyond syntax is a fruitfulventure. An interestingproblemarises, however,
when we try to identify the two parallel processes: word-internalstructure, which is
absolutely necessary for the correct operation of the Word Rule, does not count as
branchingin terms of the CompoundRule. If it did, we would find *laborunion (where
union branches, being bisyllabic) contrasting with labor day (day a monosyllable,
therefore nonbranching).Evidently, words have an integrityfrom the point of view of
phrasal stressing. The notion that is relevant to the word/compoundrule is "branches
at the same prosodic level". The notion of "level" can be markedquite effectively in
metrical structure,by stipulatingthat all words be dominatedby a special node M (for
mot, W being taken). Under this assumption, labor union has the full representation
(32):
(32)
s
lab or union
MARK
270
LIBERMAN
AND
ALAN
PRINCE
Let us provisionally accept this method of notating the concept "prosodic level".
With this modificationof the formalism, the Word Rule (28) and the CompoundRule
(8b) are seen to be one. Since compounds are dominated by lexical nodes, arising
recursively from rules such as N -- N N, we designate this uniform condition on
English stress contours the Lexical Category Prominence Rule (LCPR):
(33) LCPR
The LCPR is known to admit of various kinds and classes of exceptions. Bisyllabic
1
prefixes, especially the Greek, like main stress: monosyllable, television, agriculture,
1
(*inhibitory, as predicted), sedentary, chalcedony, Abernathy; words ending in phonetically high tense vowels spelled as in -oon, -ier, -ee, -oo, -oe displayfinal main stress,
1
which is not allowed for in the rule as stated: pontoon, pantaloon, deportee, engineer,
1
lavalier, bamboo, Tippecanoe. Some exceptions are merely sporadic: circumvent vs.
1
circumcise. In collocations, too, we find cases like fifth avenue (cf. fifth street), (the)
1
SPE, UN.
These data give valuable evidence about the ways exceptions impinge on otherwise general rules of language, and about how subregularitiesrelate to the wider
regularities they depart from. We shall examine these questions in detail when we
returnto the LCPR in order to broadenits factual coverage of the lexicon.
2.3. Finding [+stress]
ON
STRESS
AND
LINGUISTIC
RHYTHM
words. That such rules exist raises an immediate question: how (if at all) does the
operation of the rules that locate [?stress] in the phonologicalstring relate to the mode
of tree constructionpresupposedby the LCPR?To see the relation, we need an image
of the rule system that generates the arrangementof stressed and stressless syllables.
In this section, therefore, we present a version of the core processes of English
stressing, drawingparticularlyupon the proposals and observations of Ross (1972)and
Halle (1973).
We assume an SPE-like taxonomy of the vowel system, distinguishingtwo classes
with the feature [?long]. The class of underlyinglylong vowels is exemplified in the
following words:
(34) divine
obscene
vane
pounce
moon
v6te
Bermuda
point
This is, of course, the class of tense vowels and true diphthongs.
The underlyingshort vowels are found in words like these:
(35) pit
pet
pat
put
putt
pot
impudent
pong
Note that phonetically long [a] of pot-generally spelled o-is underlyinglyshort. The
u-vowel of Bermuda, reduced in impudent, may be either long or short underlyingly.
(For extensive justification of this type of analysis, see SPE.)
With this distinction in mind, consider the placement of stress in the following
words (which we offer as a characteristicsample):
(36) a.
America
canonical
Everest
asparagus
poly'gamous
elephant
b.
aroma
Cardona
hormonal
horizon
desirous
adjacent
c.
def6ctive
referendum
amailgam
erector
anarthrous
Charybdis
The words in columns (36b) and (36c) are stressed penultimately,those in column
(36a) antepenultimately.The distributionof stresses correlates with the shape of the
penultimate syllable. If it is "light"-i.e. if the penultimate vowel is short, and
followed by (at most) one consonant-stress falls on the preceding syllable, as in
column (36a). If the penultimatesyllable is "heavy"-i.e. if the penultimatevowel is
long (column (36b)), or followed by two (or more) consonants (column (36c)}-then it
must bear stress itself.
In all the examples so far considered, the final vowel is short. When long, it must
272
MARK
LIBERMAN
AND
ALAN
PRINCE
anecdote
execrate
ballyhoo
complete
divine
devote
mahout
exploit
The rule is intended to apply in accord with the principleof disjunction,so that to
any given word only the longest applicablesubrulemay apply. The sense of this is that
one must "skip over" the last two syllables if the end of the word can be analyzed by
the formula-CO V C V Co#; that one must skip over the last syllable if it has the shape
-CO V Co# (i.e. contains a short vowel); that in the other cases stress must be final.
Notice that the penultimateterm in the stress rule-(V(C)-has been generalized
beyond the data of table (36) to allow for sequences of vowels. Such do indeed occur in
words like alien, simultaneous, radium, labia, but with a wrinkle: the first vowel of the
sequence, lying in the word's penult, is phonetically tense, and would thus seem to
controvert the V requirementof our rule. As it happens, however, phonetically lax
vowels are barredfrom prevocalic position in English; there can be no words such as
*[dIdn], *[lean]. We assume, then, that the surface [iy] of words like alien is
underlyinglyshort, and that it is lengthened (or perhaps merely tensed) by a rule that
expresses an exceptionless generalizationabout English sound structure:
(39) Vowel before Vowel
V-+[+long]/
Similar to alien, etc., are cases like potato, albedo, spumoni, Mary (cf. Marie),
where an evidently tense final vowel receives no stress. Here again there is a gap in the
phonetic surface (of most dialects): no nonlow vowels appearlax finally. Reservingthe
low (front)vowel as the source for final schwa in such words as algebra (cf. algebraic),
9 It appears likely that the stress and stress-relatedphenomenawe will discuss ought actually to be
analyzedin terms of syllable structure.Whena sequence V C C V C0# is divided V-C C VCo# by the rules
of syllabification,it is usually skipped over, even though it cannot be analyzed by the formula we have
written, (V (C)) (V CO)#.Thus: al-ge-bra, pe-de-stal, re-gi-strant,in-te-gral. However, we will retain the
familiar segment-concatenatingnotation of SPE, because the details of syllabic representationare not
directly germane to the issues we are exploring in this article. The reader is invited to give a liberal
interpretationto the recurrentformula-V (C)- and indulgentlyread it as "light syllable", just as we shall
indulgentlyrefer to it by that name. An importantdiscussion of syllable structureand its role in English
phonologyis found in Kahn (1976).
ON
STRESS
AND
LINGUISTIC
RHYTHM
we shall assume that the following rule lengthens the underlyinglyshort vowels at the
end of potato, etc.
(40) Word Final Lengthening
V
#
[+long]I
[-low]
The two rules can be combined, using the ordinaryresources of generative phonology,
into the following expression:
(41) Exceptionless Vowel Lengthening
V
-*[+long]I
(V X)b#
[(+low)a]
Condition:a D b
Rule (41) predicts that ceteris paribus we should find evidence of an underlying
vowel-length contrast that affects stress and its concomitants, but that is not directly
mirroredin phonetic vowel tenseness. And indeed we do find Ohio contrastingwith
radio, and veto' (with its aspirate [th]) contrasting with mott6 (where the flap [D]
indicates a stressless final syllable, as Kahn (1976) demonstrates).
Rule (41) is ordered after the stress rule, of course, so that the underlying
distinction it masks may nonetheless have its effect.
At this point it is worth noting that a numberof words show some degree of final
stress even though they terminate in the sequence -CoV Co#: for example, Aztec,
insect, Arbib, pyramid, electron, Vermont, Berlin, chaos. Ross (1972) achieves a fair
274
MARK
LIBERMAN
AND
ALAN
PRINCE
one or another subrule, with the effect that the full count authorizedby the ESR is not
always permitted.
As we turn to examine the location of stressed syllables in regions beyond the
domain of rule (38), we find that, if we measurefrom the stress laid down by the ESR,
there are three modes of determiningthe next stress, with choice of mode governed
generally by morphology, occasionally by lexical item. For reasons that will become
apparent,we will designate these modes by the mnemonic labels the weak, the strong,
and the long.
2.3.1. The Weak Mode A number of affixes that bear stress behave like -oid with
respect to placement of stress. The diphthong[3y] patterns with the other diphthongs,
so we shall assume it has a long nucleus underlyingly;it will thereforereceive stress by
the ordinary operation of the ESR. Placement of stress elsewhere in the word is
controlled, apparently,by a kind of repriseof the light-penultcondition, as these words
illustrate:
(42) a. pyramidoid
encephaloid
tentaculoid
cartilaginoid
b. hominoid
crystalloid
an'thropoid
solenoid
c. ellipsoid
molluscoid
cylindroid
salamaindroid
d. lithoid
ovoid
theroid
centroid
Notice that the markedstresses occur exactly one syllable away from -oid in columns
(42a) and (42b), where the syllable skipped over is light (hominoid);in column (42c)
where the relevant syllable is heavy, it receives the stress (ellipsoid); in column (42d),
we see that a solitary syllable preceding -oid will also be stressed, light or heavy
(lithoid).
CO(V (C))
[+ stress]
Like -oid are affixes such as the following:
(44) Weak Retracting Affixes
-ite
-on
-ode
-ide
molybdenite
selenite
cellulite
dynamite
positron
baryon
palinode
stalagmite
stalactite
archimandrite
gelignite
electron
cyanide
telluride
peroxide
electrode
samite
lepton
hadron
cathode
epode
nitride
ON
-ology
AND
LINGUISTIC
alumni
Gemini
-i
STRESS
phenomenology
RHYTHM
foci
delicti
arachnology
odontology
egyptology
Also in the same category are monomorphemicwords like Adirondacks, Mas(cf. Ross (1972, 285)), whose final syllables bear stress
somewhat idiosyncratically,but bear it nonetheless; and Ore'stes, Laerte's, Di6genes,
(with a weak penult); and similarly Ulysses, Achilles, with an exceptionality exactly
parallelto that of bacillus, Agrippa.
sapequod, Agamemnon
[+stress]/
CO(V CO) V
[+stress]
The following table provides a characteristicsample of the forms that fall in the domain
of the rule:
(46) Strong Retracting Forms
-ate
manipulate
articulate
originate
hydrogenate
syllabicate
humiliate
salivate
defecate
confiscate
designate
exacerbate
rotate
orate
adversary
sedentary
voluntary
momentary
salivary
infantile
mercantile
byzantine
argentine
officiate
misc.
assassinate
anecdote
palindrome
cavalcade
Arkansas
nightingale
caterwaul
recognize
recognition
surreptitious
276
MARK
LIBERMAN
AND
ALAN
PRINCE
Notice that the forms salivate, defecate, because of their relation to saliva, feces,
plausibly have a long vowel underlyinglyin their medial syllables, supportingthe claim
that any syllable may be bypassed.
The forms cited suggest the sometimes vacillant character of the distinction
between strong and weak retraction. Although most -ate verbs go by the strong
principle, there is a smattering of those that do not; e.g. impregnate, infiltrate (to
some), demarcate, incarnate. These we will regard as exceptions to the lexical
redundancythat assigns -ate verbs to the strong class. As for the forms in -ation, there
is a well-known class of apparent counterexamples to the claim that it is a strong
retracter,typified by examples like relaxation, exaltation; these, however, will find an
explanation below when we turn to the phonological cycle. The cited forms in -ary
stand in contrast to others like elementary, anniversary, infirmary (cf. Halle and
Keyser (1971, 41)). A word like prehensile retracts weakly, as opposed to infantile and
mercantile; as against byzantine and libertine stand benedictine and elephantine. The
leftmost of these two syllables must be light; otherwise, we find, for example,
Monongdhela, never *M6n6ngdhela.These facts indicate the following formulation:
(47) Long Retraction
V
[+stress] I
[+ stress]
Words subject to Long Retractiondivide into five basic classes:
(I) Monomorphemicforms in which the sequence #C V C V C V precedes the
stress laid down by the ESR (e.g. catamaracn). The effect of the rule is to supply initial
stress. It seems that binary alternationof stresses is preferred if four light syllables
precede the ESR stress: consider the word (authentic)P6pocatepetl.
(II) Words in which the sequence V V immediatelyprecedes the syllable stressed
by the ESR, e.g. toredd6r, alfenate, ideJl6gical. To this class are assigned all forms
with the relevant structuralproperty, regardless of whether their morphology would
normallyput them in another class; so we have meteoroid, metM6rite (from the weak
class), and allinate, deterOradte(from the strong class).
(III) Greek prefixes like helico, hetero, as in heterodox, heteronym (cf. heteronymous), helicograph (helicography). For some discussion, see Siegel (1974).
ON
STRESS
AND
LINGUISTIC
RHYTHM
Monongahela
deteriorate
III. Greek
IV. -atory
V. miscellaneous
etoilate
ameliorate
orientate
alienate
ideological
meteoroid
meteorite
heterological
heteronym
heterodox
helicograph
automobile
hallucinatory
manipulatory
articulatory
peregrinate
disciplinary
veterinary
oxygenate
laryngoscope
compensatory
confiscatory
condemnatory
2.3.4. Retraction There is remarkable similarity among the three rules we have
278
MARK
LIBERMAN
AND
ALAN
PRINCE
distinguished,in both form and function. They all serve to place a stress to the left of
the one laid down by the ESR; two of them, the Weak and the Long, depend crucially
on the notion "light syllable"; and the Strong Rule, though independent of syllable
weight in its operation, retracts stress to a maximal distance of one syllable, just like
the Weak Rule.
We take it as significant,then, that the Weak Rule and the Strong Rule are actually
subcases of the Long Rule. We can therefore state what is basically the Long Rule as
our one rule of retraction, indexing the parentheses for ease of reference to the
subcases:
(49) Stress Retraction Rule (SRR)
V
[+stress] /
Co (V (C))a (V CO)b
[+stress]
The expansion (-a), i.e. term (a) null, is rule (41), Strong Retraction;the expansion
(-b), term (b) null, is just rule (39), Weak Retraction.We shall assume that words are
markedin the lexicon for which of the three cases-(-a), (-b), unrestricted-provides
their stress, and furtherthat these marksare distributed,whenever possible, according
to morphologicaland phonologicalsubregularitiesof the type we have been surveying.
The Stress RetractionRule is strikinglysimilarto the ESR, which is repeated for
convenience of comparison.
(50) ESR (= (38))
V -- [+stress]
/I_
Both rules measure leftwardfrom a fixed point of reference, the ESR from a word
boundary, the SRR from a stressed syllable; and the standardof measure is in both
cases virtually the same. This parallelismstrongly suggests that we are witnessing a
single unified process of stress assignment,repeatingitself across the word, feeding on
its own output. We therefore represent the entire class of processes as one basic
iterative rule:"
(51) English Stress Rule, Iterative Version
V
V
[+stress] I
CO(V(C))a(
Conditions:'2 -c D d; -a,
C)b(
X)c #
(-long)d
[+stress]
-b under certain morphological and lexical
circumstances.
11 It may well be that the correct formalizationof the notion "iterative rule" is actually in terms of
simultaneousapplication. We will use iterative language throughout,however, because of its familiarity,
since nothing in our analysis decides between sophisticated versions of iterativity and simultaneity.A
simultaneous-applicationtheory of great conceptual and empirical interest, which provides a simple
formulationfor the ESR, is developed in Vergnaud(1974; 1976).
12 The expression "'c" holds when the factor labelledc is absent from the expansion of the rule; that
is, on the first iteration.The conditionrelatingc and d merely says that a word-finalsyllable must contain a
short vowel in orderto be analyzedby factorb and "skippedover" by the rule. This bringsthe presentrule
into complete accord with the preliminary,noniterative version of the ESR, (38). Of course, no such
restriction on factor b obtains duringfurther iterations, as the phenomenaof long and strong retraction
demonstrate.
ON
STRESS
AND
LINGUISTIC
RHYTHM
Observe that this rule is rather more general in its applicability than has been
implied in our discussion of particularexamples; it will mark not just the two stresses
discussed above, but will proceed on its inexorable leftward course so long as there
remains a syllabic region to traverse. And indeed, the stress pattern of longer words
affirms the iterative expectation. Examples like Popocatepetl, Agamemnon,
Massapequod, articulatory, show that there is no restriction to having only two
stressed syllables per word, and show also that the third applicationof the rule, giving
initial stress in the cases cited, follows the same principle of syllable-skippingas the
earlier applications, at least as far as the data can indicate.
There is one phenomenon, however, that appears to challenge the assertion;
namely, the failure in certain cases of initial light syllables to show up with the
predicted stress: we find, typically, p6lice, pdrade, M6nonigahe'lawith a reduced and
stressless first syllable. Heavy initials, on the other hand, show the expected stress:
M6ntana, fanidatngo, panjandrum, articulate, totality.
Taking account of this phenomenon, Halle (1973) proposes that the stress rules be
allowed to operate freely and that their excess be trimmed back by a simple rule of
destressing:
(52) Initial Destressing (-Halle's
V
[-stress] / #Co
-
[-long]
(22))
(C)
[+stress]
Although we shall find reason to modify the rule later, we accept it provisionallyhere,
since the insight it expresses will stay with us.
2.4. Relation to Tree Theory
280
MARK
LIBERMAN
AND
ALAN
PRINCE
stressing: that each time a rule applies [+stress], the syllables in the domain of that
applicationare organized metrically, in the only way they can be. Thus in a word like
teleological, the first iterationwill, by this account, give as output the following:
(53)
A\
ww
teleologi'cal
The second iterationwould produce (54):
(54)
s wws ww
(preliminaryversion)
For these cases, stipulation (18), to the effect that stressless material may not be s,
determines completely the structureand labellingof the trees.
In order to realize the more global aspects of tree geometry by iterative
construction, we need to add two further basic clauses. First, any (branching)tree
erected by the stress rule must be joined to the arborealresult of earlier applications.
At this joining, the Lexical Category Prominence Rule is to be invoked. The second
iteration on teleological would therefore actually produce the entire tree for the word:
M
(55)
w
teleological
+ -_
_-
(56)
w s
ZcAL
reconciliation
ON
STRESS
AND
LINGUISTIC
RHYTHM
So far we have simply followed the descriptionof tree shape sketched in 2.1. We
need to depart from it somewhat, however, to give an adequate account of the
bracketingof sequences of stressed syllables. In terms of applying the ESR, we must
consider the "degenerate" case in which the immediately preceding iteration fails to
erect a (branching)tree.
An example is provided by the word odontology. The first iteration stresses the
antepenult, successfully arborizing the sequence -tology. The second iteration, an
instance of weak retraction, stresses the syllable -don-; but no metrical structure is
erected. The relevant domain of the rule is but one syllable in extent; and since
metrical structure subsists on relation, none is created. Since no tree arises from the
stress rule, there is no joining with the result of previous application, i.e. the tree
above -tology. After the second iteration, then, -don- stands alone. On the third and
final iteration, however, for reasons that will become clear when we look at vowel
reduction, we want -don- to be groupedwith o. We will therefore stipulatethat at each
iteration any stray structurelessmaterialfrom the previous applicationis to be picked
up; here again the Lexical CategoryProminenceRule provides labelling.
The thrust of this is to establish binarygroupingas a kind of baseline for prosodic
organization. If the stress rule results in a sequence of contiguous stressed syllables,
+ + + + + + +, this provision will ensure, at the lowest level, a structuring
...
+ (+ +) (+ +) (+ +). Notice, too, that by the LCPR each of these units will be
(57)
Metrical Bracketing
282
MARK
LIBERMAN
AND
ALAN
PRINCE
(58) /odontology/
+W
t '
+
ESR
sA
+ +
WI
(V) #
ON
STRESS
AND
LINGUISTIC
RHYTHM
This rule, a truncated version of (41), produces V V# from the long expansion, and
V# from the short one. Because the two subrules are collapsed by parentheses, we
expect disjunctiveapplication,even though the two rules are in no sense contradictory.
But when we look to the crucial cases, words ending VV# underlyingly,we find that
both subrules apply: as in rodeo, radio.
With stress rules, construed metrically, the logic of the situationis quite different.
Rule applicationnot only marks a segment but also concomitantlyimpartsstructureto
the environs of that segment. Each subrulecorrespondsto an alternativestructuringof
the string; and each of the alternatives is intrinsically incompatible with the others,
under the minimal assumption that a given stretch of string can have only one set of
metrical relations defined on it. Conjunctiveapplicationof a set of subrules would, in
general, breed a chaos of conflictinginterpretations;consequently, something like the
theory of disjunctionis specificallyrequiredto regulatethe operationof stress rules.
It has long been known that the distribution of reduced vowels in English words
depends heavily on aspects of stress pattern and syllable structure. A major achievement of SPE was to successfully project the surface occurrence of schwa from
representationsgenerated by the battery of stress-placementrules.
Within any theory that observes a distinction between hierarchicalprominence
relations and the simple presence-absence of stress, it is possible to pose an interesting
question: can vowel reduction be keyed to just one of these two facets of representation, and if so, which one? In Halle (1973), where this distinction is upheld, the central
process of reduction (Halle's (25), the Destressing Rule) applies when only [+stress]
has been marked;and crucially so, for in his system patternsof relative prominenceare
shaped in essential respects by the effects of this destressing. In SPE, on the other
hand, there was no choice but to base the rule on a (numerical)stress-level configuration, i.e. relations of prominence. We will argue in this section that the SPE approach
is the correct one, and we will show that the generalizationsgathered in SPE can be
stated in a particularlysimple form within the present theory; indeed, we will find that
the destressing rule itself makes reference to neither stress nor prominence.
There are three basic and unmistakableenvironmentsin which vowels characteristically lose their grip on fullness. First, initial syllables lose stress when they are light
and immediately precede a more strongly stressed syllable. The following table,
13We are indebted to Debbie L. Nanni for many valuable suggestions and comments that have
improvedthe analysispresentedin this section. The readeris referredto Nanni (1976)for an interpretationof
these phenomenafrom a rigorouslysyllabicpoint of view.
284
MARK
LIBERMAN
AND
ALAN
PRINCE
Initial Stressing
a. Light
police
Monongahela
balloon
asparagus
mosquito
astronomy
b. Heavy (long V)
c. Heavy (closed)
psychology
totality
Daytona
neutrality
tisane
maintain
bandana
lactation
pontoon
sectarian
Cartesian
technique
There are some exceptions to the generalization,going both ways. A few words like
raccoon, tattoo, settee, esprit have nonreduced short vowels in initial, open syllables.
Words like ph6nology, schematic, bdnality, legality are transparently related to other
words that imply a long vowel initially:phonate, schema, banal, legal. These sporadic
weakenings we shall regardas the result of a nongeneralminor rule of shorteningthat
applies to selected morphemes. Words with closed initial syllables provide some
apparent counterexamples to our claim: e.g. Kentucky, Vermont, Berlin. Of these,
Kentucky is perhaps truly an exception, but Vermont, Berlin are phonetically quite
regular:notice that *[vermant], *[berlln] are impossible. The vowels [I], [e], [A], [a]
have in certain environmentscoalesced with [r] in standardAmerican to produce the
"r-colored" mid-central vowel; when this vowel is metrically weak and nonfinal, it
uniformly destresses. It is quite plausible to assume one of these vowels for the first
syllable of, say, Berlin; then the phonetics of the word will follow regularlyonce the
prominence contour is fixed. (See Kahn (1976) for an interestingdiscussion of related
issues.)
Leaving aside these cases, which are either sporadicor explicable phonetically,we
find ourselves beset by hundredsof apparentexceptions to this law of initial syllables
when we look into the behavior of Latinate and Celtic derivational prefixes. The
following table suggests the regularitiesof the situation:
(61)
Prefix Stressing
a. Full
condensation
advantageous
abnormality
b. Reduced
condense
advance
absurd
adeptN
adeptA
adaptation
conformation
prolongation
relayN
relaxation
pretense
adjacent
conform
prolong
relayv
relax
pretend
c. Full
contemplate
adulate
abnegation
adjuration
confirmation
product
relative
replicate
predicate
ON
STRESS
AND
LINGUISTIC
RHYTHM
MacDonald
Maclntyre
MacTavish
MacIntosh
Macllwain
MacDonough
MacBride
MacAnulty
Far from being erratic,these cases conformto a generalizationof their own that, in fact,
closely resembles the one we have just seen. Prefix syllables, light or heavy, typically
reduce when followed by a more strongly stressed syllable. The examples of column
(61a) are chosen to demonstrate that mere presence of a following [+stress] is not
sufficient to trigger reduction, i.e. that the structure of prominence is the relevant
conditioningfactor.
Here again there is a certain amount of irregularityto contend with. The prefix
trans- never reduces, althoughphonologicallysimilaritems do, for examplepro-, pre-,
ex-. In the words concave, convex, the prefix retains stress; curiously, in the
derivatives concavity, convexity it seems easily destressable. A more general and
interestingphenomenonis the widespreadfailure of reduction-to-schwain vowel-initial
prefixes like in-, ex-, ab-, ad-, etc. We have [Int6ns] not *[zntens]; [Igzxekt],not
seem possible (and are given by Kenyon
*[zgzXkt];but both [3bstrxkt] and [aebstraekt]
and Knott (1953)).
The rhythmicbehavior of these prefixes gives us an angle on the phenomenonof
nonreduction. We will see below in section 3.4.2 that, in the clear cases, prominence
never shifts under rhythmicalinfluence onto a stressless syllable:maroon sweater thus
contrasts with thirteen me'n. Similarly, words like intense, exact never undergo the
"rhythm rule": *intense light, *exact answer. We conclude that these prefixes are
indeed stressless, but that their initial vowels are immune to reduction-to-schwa,or
perhaps merely subject to special assimilations. Words like abstract do participatein
the rhythmic alternation, e.g. abstract phonology, and thus are destressed only
optionally. Notice that a word like advanced is a nonparticipant, e.g. *advanced
calculus, and is therefore always subject to destressing. The picture that emerges is
this: while in general the prefixes are obligatorily destressed in the appropriate
environment, a certain amount of word-and-morpheme-governedoptionality (e.g.
abstract) has managedto hold out in the lexicon.
The third major weakening affects vowels, long or short, that lie in medial open
syllables; in closed syllables, stressed vowels are preserved, as the following table
illustrates:
Medial Phenomena
(62)
a.
b. Medial, reduced
explain
rotate
component
parasite
define
explanation
rotatory
comp6nential
parasitology
definition
c. Medial, closed
odontology
evangelical
adventitious
Halicarnassus
relaxation
286
MARK
LIBERMAN
AND
ALAN
PRINCE
Notice that this process involves not only shorteningbut also destressing, whether
cyclic derivation is assumed or not. The suffix -ology induces weak retraction,as we
discovered above in 2.3; consequently, the last syllable of the stringparasft- will be
stressed when the ESR iterates across parasitology.
Can this rule of shortening-destressingbe stated in terms of the arrayof segmental
[?stress] marks? Not if it is to have full generality. Consider words like advisory,
persuasory, supervisory. The suffix -ory catches stress, and retains it, in words like
promissory, admonitory, articulatory. The reduced allomorph /ari/ occurs predictably
[mstress 1
--
-stIes
)c (C)
V
[n stress]
( +long)a
ON
STRESS
AND
LINGUISTIC
RHYTHM
[m stress1
[ stress]#
-ong
Co0
(C)
V
[n stress]
-long
Condition: m > n
This rule is intended to apply to words like those cited under (60a): p6lice,
M6nongahela, bdll6on, dspacragus, m6squito, dstronomy. The last three words clearly
indicate, as noticed in Kahn (1976)and Nanni (1976), that the rule ought to be stated in
terms of syllable structure, referringrather to a syllable-finalshort vowel. However,
the issue of syllabic representation,if not orthogonalto present concerns, is at least at
a steep angle to them, and, as with the stress rule, we will ask the reader to exercise a
certain discreet liberty in interpretingour notations.
The medial rule is obtained when b is not null, i.e. by assumingb. The (material)
implication "a D (b v c)" will be satisfied, no matter what value a assumes, the
consequent of the conditional being true; therefore, the rule will apply to any vowel,
long or short. The resultant rule takes the following form (we assume also -c, since
prefix facts are irrelevant):
(65) Medial Destressing (= (64), assuming b & -c)
[-stress]I#XVC0(C)
[n stress]
[m stress]
Condition: m > n
This rule will apply in such words as those cited under (62b): expldnation, rotdtory,
comp6nential, parasitology, definition.
The prefix rule is arrived at by assuming c (and suppressing term b, since the
relevant prefixes occur initially). For reasons symmetricalto those just discussed in the
medial case, this choice will ratify condition (2) regardless of the value of a; the rule
will therefore have in its domain both the short vowels of in =, ex =, ab =, ad=, mac-,
and the long vowels of pro=, pre, de=, re=, etc.
288
MARK
LIBERMAN
AND
ALAN
PRINCE
[stress],
long
->
[m stress]
o~0
co = (C)
[n stress]
Condition: m > n
This rule applies to such words as intense, exude, absurd, advance, MacDonald,
pretend, profound, desire, revenge.
(68) a.
b.
hAi
ws Ws
I
I1/I
+
monongahela
+
c.
+-
s ww
I I
c ntractuat
+
_-_
wA
s Wsw
IIA
I
explanation
+
++-_
ON
STRESS
AND
LINGUISTIC
RHYTHM
This point is well worth pursuing,but before we do so, we should dispel a certain
empirical cloud that hangs over the system as presently constituted. The problem is
this: bisyllabic words do not behave as we predict they should. On our account, the
ESR will always stress the first syllable; and the Lexical Category ProminenceRule
will always assign initial main stress, since a bisyllable evokes a simple two-branch
tree, the right node of which is perforceunitary. But many bisyllables are end-stressed:
police, patrol, maintain, advance, absurd, etc.
however, there does remain a residue of true counterexamples, like manure, police,
lament, which are unexplainedby anythingwe have seen so far.
Surveying the factors that influence the stressing of bisyllables, R. T. Oehrle
(1971) suggests that words of the form light syllable-heavy syllable are regularly
endstressed, and that they contrast in this with other phonological types, e.g. argyle,
microbe, topaz, gargoyle, protein, all with a heavy initial syllable.
LCPR, will then correctly produce microbe and, of course, mdnure, the only possible
outcome, the first syllable ma- having suffered prior demotion from the ranks of
potential stress-bearers.
Halle's system thus elegantly explains a whole class of apparentcounterexamples
to the LCPR in terms of an independently needed process of vowel reduction.
However, we have already encounteredthe data that belie the basic assumptionof his
theory: that destressing precedes, and thereby determines, prominenceassignment.As
we learnedfrom examples like advisory,persuasory, medialdestressing respondsto the
weak-strong distinction; prefix destressing likewise, as is shown by the contrast
between words like condensation, relaxation and words like c6mpensatory, rJlax,
290
MARK
LIBERMAN
AND
ALAN
PRINCE
Danube, essay. Prefix vowels, even in open syllables, go by the law of prominence,
and can withstand the presence of a following stress: cf. adeptN, adaptation
(*ddatptation).We shall conclude, then, in the spirit of Zonneveld (1976), that the
LCPR must be amended to supply words like manure with an iambic rhythm,and that
the processes of reductionin English are based on the organizationof prominence.
In order to express the appropriateconditioning environment of the Destressing
Rule (67), we must, it seems, be able to refer to aspects of tree form in phonological
rules, rules that make adjustmentsat the segmental level, based largely on features of
segmental or syllabic structure.The general question posed by our empiricalresult is
this: how much of the complex informationencoded in metrical trees is accessible to
segmental rules? In the particularcase at hand, the informationrequiredis essentially
trivial, from our point of view. Since, by principle (18), metrical s must dominate
[+stress], the rule could hardly function to reduce metricallystrong vowels, thereby
producingill-formedrepresentations.There is no reason, then, that a specific rule such
as (67) should have to refer to a metricalpropertythat follows from generalprinciples.
We need to bridge the gap between well-formednessconditions like (18) and the theory
of rule application,which interpretsthe meaningof structuraldescriptions. To do this,
we suggest the following ratherminimalcondition:
(69) No rule may apply so as to produce an ill-formedrepresentation.
(Notice that this is not tautological: the notion of well-formedness could be relevant
only at the underlyinglevel; or we could permitrules to deform structurearbitrarily,so
long as all damage was repairedbefore the surface.)
With principle (69) constraining applicability, the redundant elements can be
removed from the statement of the destressing rule, yielding the following formulation:
(70) English Destressing Rule (Final Version)
V
-stresslI#
(X V)b CO
C=
)C (C) V
[(+long)a]
Condition: a D (b v c)
ON
STRESS
AND
LINGUISTIC
RHYTHM
The metrical version of English Destressing is not only notationally simpler than the
SPE rule; it is also somewhat more generalin its applicability.Considera structurelike
the following:
R
(71)
w wI
WI
Vl CoV2C
+
V3
_
s
-long
]/ [1 stress] Co__
CV
Words like classificatory, significatory, respiratory (as opposed to divinatory) will not be treated as
centraldata here.
292
MARK
LIBERMAN
AND
ALAN
PRINCE
(71) is appropriateto those cases where the rule of Poststress Reduction is called on,
then our metricized version of English Destressing, (70), will cover these new data
without furtherado. This would be a real result for metricaltheory; a w is a w, whether
it lies to the left or to the rightof its strong sister, and it is difficultto see how this kind
of equivalence between pre- and poststress positions could be capturedin a theory that
does not make use of constituent structurein the way we are proposing. Notice, too,
that it is not a question of collapsing a new rule with the old one by means of a
notational convention. The old rule, in its very simplest form, predicts destressing in
the new environment;it would have to be complicatedif it were not to apply.
The primarymotivation for Poststress Reduction comes from the behavior of the
suffixes -ory, -ary. As noted above, these suffixes receive stress by the ESR: e.g.
promissory, laudatory, sedentary, preliminary. We will therefore assume, with SPE, an
underlying long vowel, giving -ory, -airy; the lax quality of the surface vowels is a
(73)
As
ww
I I
cursory
However. if we start out from the expected base form, curs+o6ry,applying the ESR
and the LCPR will confer the following incorrect tree upon the word:
M
(74)
wAw
I
I I
cursory
Figure (74) highlights the second peculiarity of the suffixes, alluded to briefly
above in section 2.2: when they end a word, they never bear main stress, even though
the LCPR predicts that they should. We might consider them simply exceptional to the
LCPR, or somehow lexically marked for weakness, so that the interpretationof
cursory would be like this:
RHYTHM
(75)
s w
*cursory
ESR
ESR
Alternation Provision, LCPR
M
o
[ -long
EDR
1
-stress]
Output
I I
cursory
294
MARK
LIBERMAN
AND
ALAN
PRINCE
(Of course, further rules will apply, e.g. the one reducing all stressless lax vowels to
schwa, the one assimilatingvowels to Inr,etc.)
Because -ory is metrically unitary rather than branching, it is assigned the label
weak by the ordinaryaction of the LCPR;because the first vowel of the suffix is weak,
and because it is followed by the sequence -CV, it will reduce by the English
Destressing Rule. Thus both peculiaritiesfollow from the extrametricalityof the ending
-y.
If the sense of the present theory is that stress provokes the organization of
syllables into constituent groupings,it is reasonable to assume that every syllable is to
be accounted for metrically.We need, therefore, a rule (or generalconvention)that will
join the wanderingsyllable -y to the tree of the word it belongs to. We know of little
evidence that bears on the details of this adjunction,but we will propose to accomplish
it in a "structure-preserving"fashion, that is, so that the output resembles a tree that
would have been generatedunder more normalcircumstances.
To this end, let us define a notion left foot.
(77) Definition: Left Foot
We can now frame the convention for adjunctionin terms of this notion.
(78) Stray Syllable Adjunction (SSA)
A syll
M(F)
M(F)
s w
s ww
I I
I
IcI
cursory =>cursory
ON
STRESS
AND
LINGUISTIC
RHYTHM
b.
M
s(F)
,jF)
Yv
Is t
dispensary
C.
s
I> w
dispensary
M
M
s(F)
wAs(F)
The nodes that play the role of "nearest maximalleft foot" have been markedwith (F)
to clarify the procedure.
Stray Syllable Adjunctionis evidently active in the derivationof words other than
those of the -oryl-aryclass. There are in English a few words like ancestor, carbuncle,
autopsy, necropsy, ancestry, which on the surface appearto contravene the LCPR. It
is strikingthat words of this class tend to end in sonorants or orthographic-y. If these
are underlyinglynonsyllabic, the general functioningof our rules, with some Rossian
concern for stressing final syllables -cestr, -uncl, etc., will give the correct prominence
pattern. Syllabification of sonorants in the environment C
__
for Stray Syllable Adjunction. Words like curmudgeon, Carmichael, Hoboken should
also be treated this way; note the characteristic final sonorant. These words are
exceptions to the Destressing Rule, which should apply to reduce the penultimate
vowel; however, they are exceptional in any theory that regularlyreduces -ory in such
words as cursory-perhaps the rule syllabifyingsonorantsm, n, 1, r is orderedafter the
EDR. (Notice, though, that there are analogous exceptions among trisyllables of the
-ary class: primary, rosary, binary, library, contrary. These contrast with the regular
plenary, granary, rosary, rotary, pessary, peccary, etc.).
Words in -atory-and their underived analogues like alligator, caterpillar-present
an additional problem of analysis. Consider how the present theory would treat
articulatory: the ESR applies three times, stressing -or-, then (by long retraction,
typical of -atory) -tic-, and finally, of course, the first syllable. This will result in the
tree representedin (80a) below, which is adjustedby Stray Syllable Adjunctionto that
of (80b).
296
MARK
LIBERMAN
(80) a.
b.
AND
ALAN
PRINCE
l\s
SSA~~~~
I Plww
articulatory
lWffWTI
articulatory
N
s1 \
\
A2
N
2
A1 A2
A3
A2 A3
Observe that A2 will have to be [+stress] for the rule to apply, by general convention,
and we do not need to stipulate this restrictionin the rule itself.
Foot Formationwill have the following effect on a word like articulatory:
ON
(82)
LINGUISTIC
RHYTHM
W SFWWW
S WWS
WW
!I I Ii
I
I i I II
W
I
articulatory
articulatory
+--?
AND
STRESS
-+-
Notice that Foot Formation does not provoke a revaluation of prominence by the
LCPR.
Once Foot Formation has made its emendations, the English Destressing Rule can
freely apply to reduce all medial weak vowels that lie in open syllables, the /l/ of
contradict6ry as well as the a/e of expldnation.
These two rules of readjustment will combine to ensure correct derivations for a
large number of words from the learned vocabulary, latinate and greek, in which the
phenomenon of extrametricality skews the expected pattern of relative prominence. So:
from -ory, we have promissory, category, allegory, etc.; from -ary, sanguinary,
sedentary, emissary, dignitary, and many others; miscellaneously from the -y class,
sanctimony, parsimony, narcolepsy, chalcedony. Outside of these, there are quite a
few words ending in syllabic liquids that display the same prominence (and reduction)
properties: caterpillar, pumpernickel, alligator, alabaster, paradiddle, salamander,
participle, Aristotle, axolotl. If words like these end lexically in nonsyllabic /r,lI, then a
simple and exceptionless rule syllabifying sonorants in the environment C __#,
ordered anywhere after the ESR, will, given the readjustments necessary to model the
-oryl-atory contrast, produce exactly the observed surface forms.
Our decision to maintain the simplicity of the EDR and the LCPR by treating
certain final syllables as "extrametrical" (or underlyingly nonsyllabic) has a certain
apparent cost attached to it: the rules of readjustment that we called Stray Syllable
Adjunction and Foot Formation must be spelled out. Of these, the first, or something
like it, is clearly necessary in any grammar that has syllables created in the course of
derivation, and is thus a prime candidate for the status of general convention, rather
than grammatical rule. The second is perhaps more specifically motivated, but its
effects, too, are quite natural, and it could easily be regarded as a general law rather
than a feature of English. If this is correct, then what we have presented as rules of
readjustment ought rather to be seen as consequences of the definition of tree form in
MARK
298
AND
LIBERMAN
ALAN
PRINCE
There is a strikingdifference in kind between the evidence for the phonological cycle
within words and the evidence for the cyclicity of phrasalstress rules. The motive and
the cue for the phrasal cycle has been the fact that, in languageslike English, patterns
of relative prominenceare largelydeterminedby syntactic constituent structureand are
quite generally preserved under embedding. Within words, however, such motivation
from the character of prominence phenomena is entirely lacking. Morphology, the
analogueof syntax, falls far short of providinga constituent structureof syllables that is
adequate to metrical labelling. On the one hand, many polysyllables are monomorphemic, (e.g. Tatamagouchi), so that morphology has nothing to say about their
internalorganization;on the other hand, when morphologydoes provide structure,it is
typically irrelevant to metrical grouping. Consider, for example, the word compensation: morphology will analyze it as [[compensat] ion], while phonology must see the
principal significant division as [[compen][sation]]; the two parsings are grossly
incompatible.
In addition, morphological embedding freely disrupts the pattern of relative
prominence. In compensate, the first syllable predominatesover the third;in compensation, the relationship is reversed; in compensatory, these two prominences are
utterly reduced, and the weakest syllable of the inner constituent (-pen-) becomes the
strongest syllable of the entire word. The prosodic constituent structure varies
correspondingly,changingfrom word to word, regardlessof the constants of morphological relatedness.
M
(83)
a.
/s
M
t
[comp11 Inaatev]
| mF%nsatv]ionN1 c. [[compensatv]
b. [[c%
ryA]
ON
STRESS
AND
LINGUISTIC
RHYTHM
The primary evidence for the subword cycle comes from the persistence of the
segmental mark [+stress], unmooredfrom any syntagmaticrelations it may enter into.
Consider the following arrayof facts:
(84) a. Base TypeI
advantage
impregnate
infest
subjective
abnormal
indent
report
b. Embedded
advantageous
impregnation
infestation
subjectivity
abnormality
indentation
reportorial
c. TypeII
compensate
designate
orchestrate
anecdote
demonstrate
concentrate
recognize
d. Embedded
compensation
designation
orchestration
anecdotal
demonstration
concentration
recognition
The interesting action takes place in the second syllable. In column (84b), the
second-syllable vowels are all nonreduced [ae, c, :]; in column (84d), which contains
words of similarmake-up, in terms of segments and prominence,the analogousvowels
are uniformlyreduced; they are schwa (with perhaps some environmentalcoloration).
Whether the vowel reduces or not in the complex word correlates perfectly with its
status in the base forms listed in columns (84a) and (84c). If it is unstressed in the base,
as in column (84c), it is unstressed when embedded; if stressed as in (84a), it shows
stress when embedded, and does not, therefore, admit of reductionto schwa. (Notice,
too, a clearly perceptible rhythmicdifference between the words of columns (84b) and
(84d), a consequence, presumably,of their different(+) stress patterns.)
This kind of phonological dependency between complex words and the simpler
words they contain is widespreadin the lexicon of English and generally quite regular.
Exceptionality involving Type II words is, we believe, unknown;there are apparently
no alternations of the hypothetical form concntrate-conctntration. Among Type I
words, a certain amount of unexpected reductionis found, but it appearsto lie within a
phonetically circumscribed domain: metrically weak nonlow vowels occasionally
collapse with following tautosyllabic sonorants, even when they should bear a
protecting stress. We find such examples as commntary (commtnt), consltation
(consAlt), transftmation (transform), and, optionally, sentimntality (sentimental). Note
MARK LIBERMAN
300
b. recipr6cality
corporeality
artificiallity
orig nality
municIpality
religiosity
volu'minosity
c. Taitamagouchi
Passamaquoddy
caitamarain
heter6dyne
Winnepe-saukee
Kailamazoo
anthr6p6m6rphic
The marked vowels of column (85a) are all short underlyingly (for the last, cf.
volume), and they are stressed by the ordinaryoperation of the ESR. The interesting
contrast is between (85b) and (85c): words in (85c) have a second stress as far back
from their endmost stress as is possible; the words in (85b), which have a syllable
structureidentical in the relevant respects to that of the words in (85c), show a second
This
stress that falls one syllable short of its greatest possibilities (e.g. *orWginality).
shortfallmeans that the derived word will have a stress just where its base has one. As
with the reductioncases, we regularlyfind a stressed syllable where we could as easily
find a stressless one if the rules operatedfreely or randomly.
Paradigmslike these show conclusively that the (+) stress pattern of a complex
word depends on the (+) stress pattern that its morphologicalconstituents assume in
isolation. The most elegant and restrictive device proposed to represent this kind of
pervasive "transderivational" relationship, and the one we shall accept, is the
phonological cycle. By means of it, aspects of the derivation of subconstituents
become, literally, part of the derivation of the whole. In SPE and Halle and Keyser
(1971), the markedstresses on such words as relaxation, originalitynecessitated cyclic
applicationin a direct, "observational"way; for without it, no stress would be placed
on those syllables by the posited rule system, which differs from ours in not having an
iterative stress rule. Theories of this sort depend on derivations of the following form:
(86) Noniterative Derivation Type
[[[originN]alA]itYN]
+
+
+
originality
ON
STRESS
AND
LINGUISTIC
RHYTHM
postulated-and we have been careful to illustrate its operation at every step with
noncomplex words-has the capacity to place stress appropriately in relaxation,
originality, etc. We need simply mark such words for weak retraction. To do so,
however, would be to abandonthe generalizationthat such stress positioningcorrelates
with morphologicalcomposition. Even if it could be arguedthat weak retractionis the
"unmarked"mode for complex words, and therefore need only be stipulatedonce for
the entire class, the argumentfrom lost generalizationstill has force; with a cycle to
transmitto the whole word the features that its parts earn on their own, the fact that
suffixes like -al, -ous, -ive, -age, etc. induce weak retraction when stressed follows
directly from the fact that the ESR treats them quite normally when they end a
constituent; no lexical stipulationis required,general or specific, to guide the stressing
of such suffixes and the words they belong to.
To accommodate this generalizationwithin our analysis, we must slightly modify
the ESR and our conception of its effects. The ESR takes on the following shape:
(87) ESR (Cyclic Version)
V->[+stress]/
CO(
E
Conditions: -c D d, a
V
V
(C))a(
-long 1
[(-long)d]
- stress]
N, A, V
CO)b(VX)cJ
The rule has been changed to measure from the end of a constituent ratherthan from
the end of a word; and the term a has been further restricted so that it can only
correspondto stressless syllables. This will prevent it from skippingover the cyclically
assigned stress of originality, religiosity, etc. Term b must, however, on our account,
be able to analyze a stressed syllable, in order to correctly derive words like
compensatory.
Before applying any rules on a cycle, erase all prosodic structure in the
domain of that cycle.
This will leave the ESR with a slate that is clean except for the residue of [+stress]
marks deposited by applicationson earlier cycles.
Derivations like this will result:
(89)
ig
[[[or
it
alA]
inN]
YN]
[[comp
atev]
ens
or
YA]
'-'
+
ESR
Deforestation
ESR
+
+
w
s
/
s
VSR
Deforestation
ESR
IL L
ESR
+
M
ITs
IXw
\'
5
5
+
fi
ESR
w
T
W
wS
fff
W
Wsw
swM
M+
EDR
w
\sv
sw
ww\
\/
Output:
M
S
I I III
wrat
rig
in al i
ty
304
MARK
LIBERMAN
AND
ALAN
PRINCE
Metrical stress theory is thus brought into consistency with cyclic applicationof
rules like the ESR. (Note that the Destressing Rule (70) remains word-level.) Because
the subword cycle is not defined on metricallyrelevant bracketingthe way the phrasal
cycle is, the present theory does not illuminatethe structure-dependenceof (?)-stress
rules. As in former theories, this is a special property that does not follow from
independentassumptionsabout the nature of the grammar.
Given cyclicity, then, and our understandingof the arboreal consequences of
applying stress, a principle is requiredto adjudicateamong the conflicting structural
claims of various cyclic applicationsof the ESR; we offer (88), Deforestation, which
rules in favor of the last cycle on which the rule operates, the one that encompasses the
whole word. Further evidence is of course needed to establish (88) as the correct
interpretationof the phenomenon, in the face of the many imaginabletechnical and
conceptual alternatives. It is interesting to note that the other well-known mode of
phonological organization, grouping of segments into syllables, also changes under
morphologicalembedding:compare the second syllable of ex-plain with that of ex-plana-tion; if stress rules are sensitive to syllable structure per se, as suggested above,
then there must be de- (and re-) syllabificationat each cycle, just as there is de- and reforestation. Principle (88), then, which may be the metrical reflection of a more
embracingtheory of cyclic reorganization,allows us to present a version of the ESR,
(87), that can play a key role in representingthe system of "translexical" regularities
typified by the data discussed in this section.
2.8. The LCPR Reconsidered
the list-and it can be made much longer-one might wish to conclude that Schane's
generalizationabout main stress, which the LCPR takes over, is overblown or perhaps
entirely spurious, a mere artifactof unscrupulouslypreselected data. Such scepticism
(though it has, presumably,its uses) is quite unjustified,for an actual considerationof
the facts shows that a few subregularitiescover the bulk of apparentexceptions, and
provide a basis for understandingthe rest.
Words like pontoon, marquis, serenade belong to a sizable class of words,
identifiable by the shape of the last syllable, that have kept their end-stressingin the
passage from one side of the English Channel to the other. The following list gives
ON
STRESS
AND
LINGUISTIC
RHYTHM
many of the relevant endings, with examples to show how they attract main stress:
(90) -ier, -eer engineer, frontier, veneer, lavalier, chandelier
-oon
pantaloon, octaroon, tycoon, baboon, poltroon
-ique
unique, antique, bezique, Mozambique
-00
tattoo, kazoo, shampoo, bamboo, canoe
-ise
chemise, valise, expertise
-ade
serenade, cascade, grenade, stockade, blockade
-ette
novelette, cigarette, vignette, corvette
-ee
deportee, addressee, trustee, absentee, Tennessee
-elle
bagatelle, moselle, villanelle
-air
affair, corsair, debonair
-che
cartouche, pastiche, brioche, panache
-esce
acquiesce, recrudesce, incandesce, effervesce
-ane
mundane,transpadane,ultramontane,chicane
-ar
guitar, bizarre, cigar, (bazaar)
-eau
flambeau, tableau, chateau, portmanteau
-esque
picturesque, grotesque, statuesque, romanesque
(A useful discussion of words of this type is found in Oehrle (1971).)
Assuming that the members of this class are designated [+F], we can restate the
LCPR to take their existence into account:
(91) LCPR
long initial vowel). This process, as noted above in 2.5, is subject to a fair amount of
lexical idiosyncrasy: rabbi, satire, essay, arfstocracy (*dristocracy) should all be
iambic in the relevant metrical group. (Note that the shortness of the first vowel of
aristocracy is shown by its reduction in aristocrat.) A more interesting general
restriction is seen in data broughtto our attention by P. Kiparsky:words like lithoid,
ovoid, cathode, anode, epode, hadron, Semite, samite, Hittite are not subject to the
306
MARK
LIBERMAN
AND
ALAN
PRINCE
rule, even though they would appear to be prime candidates for it. Apparently such
affixes repel prominence, with the effect that the ESR stress on the first syllable is
allowed to surface. We can record these observations in the following clause:
(92) LCPR (cont'd)
C. N1 = # C0 V and N2 does not immediatelydominate an affix
[- long]
Note that this rule, like all rules that interact with metricalstructure,may not function
to create a strong stressless syllable; it cannot apply to words such as atom, gallop,
etc.
Verbs have their own pattern of prominence, determined partly by metrical
geometry, partly by morphology.Before we can investigate it, we shall need to make
clear two aspects of the ESR, unmentionedin our originaldiscussion, that set the stage
for the prominence rule by assigning [+stress] in places where we might otherwise not
expect it.
(1) The first iterationof the ESR may not skip over the entire stem of a word: we
find, for example, consider, infer, repel, emit, equip, discover, comment, permitN,
insect, MacDonald, MacCord, where in each case a syllable that could have been
ignored is supplied with stress. This restriction does not in general limit further
iterations, cf. concentrate (stem centr), recognize (stem cogn), MacIntyre (stem
Intyre). The notion stem here should probably be further limited to the latinate (and
goidelic) segment of the vocabulary,for the accentuationof greek words does not turn
on quite the same distinction: with them, the stem, so long as it meets the purely
phonological requirements, may be skipped over (anaLYSis, photoGRAPHy, synTHESis) when it is not absolutely word-final (photograph, orthodox, monolith). These
some others), we can regardthe essential verb rule as being the following, maximally
penultimatesubruleof the ESR:
(93) ESR for non-nominalitems, first iteration
V -* [+stress] / __
Co(V (C))a]
Notice that this is simply a somewhatattenuatedversion of the normalrule; we can say
that non-nominal items are usually (-b) on the first iteration. Exceptions to this
subregularity,like challenge, govern, bollix, modest, will go by the full ESR. Observe,
too, that as with the full ESR, we find that cases of pseudogemination-caress,
harass-can be treated as being lexically marked(-a), like Agrippa, bacillus.
ON
STRESS
AND
LINGUISTIC
RHYTHM
robust. This provision has very few exceptions (e.g. comment, ribald) and is responsible for the widespread alternation of prominence between verbs and homophonous
nouns, which go by the LCPR as originallypresented. A sample of typical participants
in the alternation:
address
transfer
permit
export
(94) survey
detail
suspect
contact
protest
insert
progress
convict
abstract
combine
record
relay
over-doseV,N,
etc. The serious exceptionality has to do not with the verbs and
descent,
nence, reminiscentof the situationin such nouns as lithoid, hadron, mightbe definable
generally on the class of affixes per se, but for a few nonproductiveexceptions (e.g.
bombard).
The third factor that influences verbal prominencepatterns is the location of the
stem: N2 is strong if it immediatelydominatesthe stem, regardlessof the shape of N1.
Typical examples are intervene, intersect, intersperse, interpose, intercept, superpose,
comprehend. Occasional exceptions are found: su'pervise, circumcise contrast with the
normal supervene, circumvent. The rule does not seem to pertain to adjectives, cf.
circumspect, derelict.
D. (Non-nominalconstituents)
(i) N1 does not branch, N2 does not directly dominate-ate, -ize.
(ii) (verbs) N2 immediatelydominatesthe stem.
We can now assemble a fairly complete picture of the prominenceregularitiesthat
308
MARK
LIBERMAN
AND
ALAN
PRINCE
obtain within English words. For concision, let us use the notation A/B to mean 'A
immediatelydominates B'. The LCPR will take the following form:
(96) Lexical Category Prominence Rule
In the configuration[N1 N2 ]
I. N2 is strong if any one of the following conditions is met:
A. N2 branches
B. N2/[+ F]
C. N1/#C0 V and not (N2/affix)
[-long]
D. a = non-nominalor [+R],
(i) N1 does not branch, and not (N21-ate,-ize)
(ii) a = verb and N2/stem.
II. Otherwise, N2 is weak
The "if and only if' of earlier statements has been dismembered for ease of
interpretation.Exceptionality has a clear meaning in terms of the format of rule (96);
words like Ladefoged can be marked as [-I], sending them straight to Rule II, the
elsewhere case, which assigns a trochaic pattern; words like advanceN, delayN,
accordN, which cling to the verbal pattern, are to be marked [+R], entitling them to
a. grass-green b. Anglo-French
c.
stone-deaf
politico-economic
crystal-clear
socio-political
skin-deep
anarcho-syndicalist
high-born d. well-meaning
ill-tempered
easy-going
clean-cut
far-reaching
hot-headed
sweet-smelling
(We are assuming that such units as politico is column (97b) count as words, and are
therefore single units in the calculationof compound prominence, being dominatedby
the node M. For a thorough and insightful survey of compounding, the reader is
referredto Marchand(1969), from which the above list was compiled.)
The compounds that do not follow this pattern-that, in fact, behave as the
original LCPR predicts-seem to uniformly have a noun as their first element: e.g.
color-blind, class-conscious,
cept for "comparative" compounds like those listed under (97a), and a few cases like
hand-picked,home-made, it appearsthat we can regardadjectivalcompounds as being
marked [-case D] when they incorporatenouns.
For verbs, the only truly productive derivational process that resembles compoundingis prefixationwith word-likeelements such as re-, de-, pre-, under, over, out,
as in re-do, de-pants, pre-date, under-cook, over-cook, out-cook. As expected, these
ON
STRESS
AND
LINGUISTIC
RHYTHM
collocations are uniformly iambic, since N1 dominates a single word and is thus
nonbranching in the relevant sense. Items like horse-whipv, stage-manage, aircondition are verbal by zero-derivation,always applicableto make a noun into a verb,
or by backformation (stage-manager, air-conditioner). Presumably, a more developed
understanding of morphology might tell us why such words slavishly follow the
prominence pattern of their sources, as in e.g. cold-shoulderN,V,mass-prodtcev (massprodLictionN), vs. h0rse-whipN,V.The cases of verbs like pistol-whip, keel-haul, spoon-
feed, which stand on their own without a nominalback-up, seem to imply that for verbs,
as for adjectives, incorporationof a noun renders a form [-case DI, i.e. subject to the
usual compound rule.
At this point, althoughdetails could be pursuedfarther,it appearssafe to conclude
that compounds behave in essential respects like other members of the lexical
categories they belong to, and that the basic branching-nonbranchingdistinction
pertains to the prosodic level where words are grouped, as well as to the subwordlevel
where syllables are hierarchicallyorganizedinto constituents.
In order to improve the LCPR to the level of adequacy represented by the
formulationin (96), we had to let it refer to morphologicaland lexical informationof a
kind that is not representedin the structureof trees. However, this informationenters
into the rule in remarkablylimitedways: it is apparentlynecessary to know only what a
given node in the formula [N1,N2a] is, as it were, to correct the errors of purely
geometric calculation. Thus, the syntactic category of a is relevant, as is, apparently,
that of N1 in verb and adjective compounds. Materialimmediately dominatedby the
N1,N2 of the formula is also pertinent under certain circumstances, but we found no
call to look further up or down the tree. This suggests a narrowly local theory of
prominence assignment, in which labelling is based only on structuralproperties of a
node (branching-nonbranching),syntactic or morphologicalproperties of a node or of
the node immediatelydominatingit (a), or on morphologicalpropertiesof immediately
dominatedmaterial.Of course, only detailed work on a range of languagescan validate
such restrictions;encouragingly,J. Bing (1976) examines Dari (Afghan Persian) stress
from a metrical point of view, presenting a thorough and persuasive analysis that
operates well within the limits we are suggesting.
We conclude, then, that the first promise of metricaltheory, to unify the rule for
prominence assignment in words and compounds, is borne out by close investigation.
Metricaltheory provides the means for a simple account not only of the broadfeatures
of the lexicon but also of the characteristiccomplexities and subregularitiesthat are so
much a part of the English stress system.
3. The Metrical Grid
3.1. The "Rhythm Rule": Description
310
AND
LIBERMAN
MARK
ALAN
PRINCE
of cases in which relative prominence is not preserved under embedding. For example,
3
he's good-looking
but good-looking lifeguard; empty bed but empty bed blues, etc.
Kiparsky (1973) attempted to formalize such rule for German, where a somewhat
2
similar phenomenon exists .16 He cites examples such as halbtot, but der halbtote
2
Mann; fing an, but er fing an zu reden. He also observes that a corresponding
phenomenon, a shift of secondary stress away from primary stress, may occur when
1
the secondary stress follows the primary one, e.g. sichtbar but unsichtbar; Grossvater
1
but Urgrossvater; anziehen but den Rock anziehen. To account for these cases, he
proposes a mirror-image rule, stating that the second subordinate stress before and
after a primary stress has [2 stress] reassigned to it, with a generalization of the stress
subordination convention to downgrade by one level all nonprimary stresses in the
cyclic domain in question.
15
These examples (translatedinto numericalrepresentation)are taken from Kenyon and Knott's (1953)
Pronouncing Dictionary of American English, which generally follows the practice of giving both stressings
ON
STRESS
AND
LINGUISTIC
RHYTHM
Note that unlike German, English does not show a symmetrical pattern of
2
rhythmic stress adjustment. Thus sports contest never becomes *sports contest; verb
2
Let us consider first how to represent the notion of "stress clash". It is not enough,
obviously, to speak of "adjacent stresses" as clashing. It is possible, indeed common,
2
for stressed syllables to stand adjacentto one another in English-in thirteen men, for
example, the syllable -teen retains some stress even though the main word stress is
retracted, producingan "alternating"pattern.
4
312
MARK
LIBERMAN
AND
ALAN
PRINCE
42
men,
Tennessee air, and achromaticlens. Suppose that the succession of syllables in these
examples is represented as a sequence of numbers, for ease in reference. Then in the
2
case of thirteen men there is another "level" of stress on which -teen and men are
represented-we symbolize this by placing a second level of marks above the first
(continuingto use consecutive integers as place-holders).The fact that men is the main
stress of the phrase can be indicated on a third level, if desired:
(98) a.
6 level 3
4
5 level 2
12
3 level l
thirteenmen
b.
6 level 3
4
5 level 2
12
3 level l
thirteenmen
We observe that (98a) has two adjacent placeholders on level 2 (labelled4 and 5)
with no interveningelements on level 1. This configurationcan be taken to representa
"clash" of stresses; and of course it is not present in (98b), where the "secondary
stress" has been retractedto the first syllable of thirteen.
In terms of the grid-likestructureswe have erected above these two stressings of
our first example, the "degree of stress" of a given syllable is representedby the height
of the column of marks that stands over it. Notice that all of the inequalitiesof stress
implicit in the pattern thirteen men (-teen >thir-, men >-teen) are captured by the grid in
(98a).
4
We now turn our attention to the second case mentionedabove, Tennessee air. A
similar attempt to capture inequalities of stress would yield the grid pattern shown
below:
(99)
8
5
6
1 2 3
10
9
7
4
level 4
level 3
level 2
level I
Tennessee air
Observe that the grid patternin (99) contains a pair of "clashing" elements", those
labelled8 and 9 on level 3. In fact, levels 2, 3, and 4 in this case have exactly the same
4
ON
STRESS
AND
LINGUISTIC
4
RHYTHM
2
Applying the same method to our third case, achromaticlens, yields a grid like
that in (100):
(100)
11
10
9
6
7 8
1 2 34 5
achromaticlens
level 4
level 3
level2
level l
Again, we see a pair of adjacentelements, this time those labelled9 and 10, which
are not separatedby an interveningelement in the next level down. There is, of course,
a syllable interveningbetween the two syllables that correspondto elements 9 and 10;
but in terms of the metricalhierarchy,on levels 2, 3, and 4 this case is exactly the same
as the two cases considered previously.
In what follows, we will suggest a more precise formalism for such metrical
hierarchies or "grids", and explore their relationship to the patterns of relative
prominence discussed in preceding sections. Once we have done this, we will be in a
better position to motivate the use, in describing English stress patterns, of this
additionalformal device.
3.3. Formalism
314
MARK
(101) a.
LIBERMAN
AND
ALAN
PRINCE
11 level 4
10 level 3
8 level2
6
1 2 3 4 5 level l
9
7
b.
11 level4
10 level 3
7 8 level2
1 2 3 4 5 level l
9
6
Terms like adjacent, alternating, clashing, etc. can now be defined in a way that
makes sense of intuitions about the role of the Rhythm Rule. Elements are metrically
adjacent if they are on the same level and no other elements of that level intervene
between them; adjacent elements are metrically alternatingif, in the next lower level,
the elements correspondingto them (if any) are not adjacent; adjacent elements are
metricallyclashing if their counterpartsone level down are adjacent. Thus the elements
labelled 9 and 10 are metrically adjacent in both of the examples given above.
However, in (101b) elements 9 and 10 are alternating,since 6 and 8 are not adjacent,
while in (lOla) elements 9 and 10 are clashing, since 7 and 8 are adjacent.
We could, of course, project the objects represented in (101) onto a single row
of symbols, by allowingthe symbols in that row to define column height. However, any
definition of alternating and clashing in terms of such a linear projection of the grid
would in essence depend on some more-or-less overt process of inverting the
projection.
Before returning to the linguistic relevance of metrical hierarchies, it will be
convenient to define some additionalterms with which to describe their structure.In a
given metricalgrid, the numberof levels is finite, and thereforethe "column" in which
any given element stands has both a bottom and a top. Let us define a function B that
maps every element of a metrical grid onto the element at the bottom of its column,
B(A) being called the "terminalcounterpart"of A; and a function T that maps every
element of a metricalgrid onto the element at the top of its column, T(A) being called
the "highest counterpart"of A. We now define the "terminalset" of a given metrical
grid as the range of the function B defined on its elements. If we define a function L
that maps every element of a metricalgrid onto the level of which it is a member, then
for any pair of elements El and E2, El is "metricallystronger"than E2 iff L(T(E1)) >
L(T(E2)).
Notice, incidentally, that as we have formalizedthe concept of a metricalgrid, the
elements of a grid's terminal set need not all be members of the same level-for
example, a configurationsuch as this is possible:
(102) a. 5
12 34
b. 5
6 7
12 3
In this case, the terminalset is {1,2,3,4,7}. Since we have defined left-right order
ON
STRESS
AND
LINGUISTIC
RHYTHM
only within a given level, no orderingis defined on the elements labelled4 and 7, and
the two graphicalrepresentationsgiven in (102) are equivalent. But the usefulness of a
metrical hierarchy depends on the possibility of mapping its terminal set onto the
elements of a temporally-orderedstring;and once this mappinghas been accomplished,
the terminalset is of course well-ordered.
Formally, this amounts to saying that for every metrical hierarchy G there is an
ordered set S and a function C that maps every element of the terminalset of G onto
some element of S, in a way that preserves ordering relations. Given this addition,
ordering ambiguities like those of the terminal set in (102) are not pernicious to our
enterprise, and indeed there may be some value in allowinggrids of this sort. However,
the empiricaldifferences are rathersmall, and it will be expositionallymore straightforward to insist for now that the terminal set of a grid be simply its lowest level. This
stipulationis most easily accomplishedby eliminatingthe parenthesizedmaterialin the
definitiongiven previously of the translevel counterpartfunction F, so that the domain
of F includes every element of all levels above the first one, ratherthan a subset of the
elements of these levels.
3.4. The "Scansion" of Speech
3.4.1. The Relative Prominence Projection Rule
10
8
7
6
1 2 34 5
achromaticlens
1
b.
9
10
6
7 8
1 2 34 5
achromaticlens
We are assuming that the elements of a grid's terminalset correspond one-to-one
to syllables; and we have previously defined "relative metrical strength" with respect
to elements of a metrical hierarchy. It follows naturally, therefore, that the "relative
strength" of the elements of a grid's terminalset ought to be congruentto the "relative
strength" of the syllables they correspondto.
This simple principle, which we will call the Relative ProminenceProjectionRule,
is almost all that is needed to ensure the correct "scansion" of linguistic material. A
316
MARK
LIBERMAN
AND
ALAN
PRINCE
Rhythm Rule operates, the grid alignments(or "scansions") that the RPPR admits as
well-formedon the basis of the input relative prominencepatterncontain stress clashes
that disappearin the scansions appropriateto the output stress pattern. For example:
(105) a. input scansion:
b. output scansion:
6
6
*4 *5
1 2
3
thirteen men
w s
12
3
thirteenmen
w
V/V
w
R
a.
(106) input scansion:
10
*8 *9
5
*6 *7
1 2 3 4
Tennessee air
5
w s
R
b. output scansion:
7
5
6
1 2 3 4
Tennessee air
s w w s
I II\
w~~~~~~~
w~~~~~~
R
ON STRESS
AND LINGUISTIC
b. output scansion:
11
*9 *10
8
7
2 34 5
6
1
RHYTHM
11
10
7
8
2 34 5
9
6
1
achromatic lens
achromaticlens
SW
/s
\VVI
w
s w
\V/V w
w
R
8
7
(108)
6
1 2 3 4
Montana cowboy
w sw s w
s
\s
w
note that a case like Montana cowboy, which fits the structural description of the
Rhythm Rule as given by Kiparsky, does not requirea "clashing" scansion, according
to our principles. In general, such forms18 are much less likely to undergo stress
retraction than the other cases we have discussed. Indeed, many speakers (including
ourselves, when we first considered the matter) feel that retraction is impossible in
such cases. However, we have noticed a few instances of retractionin examples of this
2
form, in our own speech as well as that of others. D. Jones (1956) cites Salvation
4
18
I.e. words of the form SI S2 S3. When a word boundary, single or double, intervenes between SI and
2
S2, the situation is quite different, and retraction is preferred. We have already cited good-looking lifeguard.
Among single word boundary cases, compare productive uses of affixes (where # remains) with more
2
"frozen" forms: pretonic lengthening but not *prevailing winds; invalid sample but not *insipid coffee.
Retraction is also common in cases where the stem is a bound morpheme, but the prefix is "alive" in the
2
sense that its root meaning is an important part of the meaning of the word: preceding message, interior
decorator. Note, too, the behavior of already, which is transparently a fusion of two words, even though
2
318
MARK
LIBERMAN
AND
ALAN
PRINCE
Army, a form that strikes us as possible, but not nearly as likely as the other kinds of
retraction we have discussed. After examining the nature of the retraction process
itself, we will reconsider some of the details of its distributionin speech.
The nature of the rule that, in our system, accomplishes the stress retractionswe
have been studying is indicated by the output scansions we have given in (105)-(107).
For example, in (107b), clashes have been eliminated on two levels simultaneously,
with concomitantloss of one grid element. We are assumingthat the RhythmRule does
not function simply to shift an element over in a metrical grid, which would have
producedthe pattern in (109) below:
10
(109)
5
1
*6 *7
2 3 4
Tennessee air
Instead, Tennessee (in this position) has been given the scansion appropriateto a word
like institute. Althoughthe last syllable of institute is stressed, it is assigned the relative
prominence pattern given in (1 lOa) below, so that the RPPR suggests the grid
alignmentgiven in (1 lOb):
(110) a.
M
sI
sww
institlute
b. 4
1 23
institute
In other words, the stress clashes that motivate the Rhythm Rule's application
are to be found in the metrical grid, but the rule itself operates on the relative
prominencepattern, reversing the strong-weak relation within one of its constituents.
Of course, this reversal compels a new scansion, in accord with the provisions of the
Relative Prominence Projection Rule, which, being a well-formedness condition, is
constantly applicable.
Note that this realizes the goal we described at the beginning of this section:
"First, we need an account of linguistic rhythm in terms of which the appropriate
RHYTHM
stress configurations are marked as 'clashing', thus producing a pressure for change in
stress-pattern. And second, we need a specification of the circumstances in which a
given language grants permission for such a change to occur." As promised, the
pressure is apt to arise in any language that has a counterpart to the Relative
Prominence Projection Rule, that is, in any stress-timed language. The permission,
specific to English, can be stated as follows:
(1 1) Iambic Reversal (optional)
Es
>s sw
1 2
12
Conditions: 1. Constituent 2 does not contain the designated terminal
element of an intonationalphrase.
2. Constituent 1 is not an unstressed syllable.
Condition 2 does not really need to be stated here, since rules are not allowed to
create the configurations/[-stress] as a matterof generalprinciple. Condition1 presents
the rule from shifting the "nuclear syllable" of an intonational phrase; without this
limitation, it is interesting to note, the unbridled pursuit of trochaic fluency would
drastically change the meaning of many utterances. However, even if the shift would
take place within a single word, the nuclear syllable is generally exempt from being
affected by this rule, so that an expert on the Chinese languageis a Chinese-expert,not
a Chinese-expert,even though a stress clash remains, in our terms. Likewise someone
who rides a kangaroois a kangaroo-rider,not a kangaroo-rider.However, if the main
phrase stress is removed from the word kangaroo in this last case, as in the phrase a
kangaroo-rider's saddle, "iambic reversal" becomes possible. Thus, both of the
13
12
10
1 23
45
sw
VV~~~V/
I
S
w
320
MARK
b.
1 2 3
AND
ALAN
PRINCE
14
13
11
12
9 10
LIBERMAN
45
kangaroorider's saddle
w I
vIs
w
R
"quite unlikely." For example, Marcel Proust is very strongly preferred to the
2
unretractedcase, butMarcel's book andMarcel's book seem almost equally likely, and
4
Marcel's finished
Why these differences? Given the system we have proposed, many of them are
quite reasonable. Recall that Iambic Reversal is an optional rule, which nowhere
mentions the notion "stress clash", but simply generates an optional relative prominence pattern in a wide variety of cases. In some of these, one option requires a
scansion that contains a stress clash, while the other does not. Nevertheless, in all such
4
cases both pronunciationsremain possible-one can always say e.g. thirteen men, at
the expense of lengthening the syllable -teen to compensate for the lack of an
intervening syllable. Since this lengthening seems awkward and unfluent in such a
tightly bound constituent, the alternativewill naturallybe preferred. If, however, the
lefthand element in the clash falls at the end of a constituent that is somewhat more
loosely bound to what follows, a longer "pseudopause" (lengtheningof the terminal
syllable(s) of the phrase) is likely to employed in any case to mark the constituent
boundary. In this situation, the clashing patternis not such a liability.
Thus the acceptabilityof a stress clash depends, in part, on the extent to which the
interstress intervalis prone to be lengthened,independently,by prepausallengthening.
It also seems to depend, in some cases, on other durationalpropertiesof the interstress
material, although the differences here are perhaps too delicate to be reliably investigated by means of intuitive data.
ON
STRESS
AND
LINGUISTIC
RHYTHM
In cases like Montana cowboy, Salvation Army our theory predicts no strong
pressure for retraction, since the original stressing produces no metricalclash. Iambic
Reversal is possible here, as it always is, but the unretractedcase is not metrically
awkward,accordingto our analysis, and thereforeretractionshould be much less likely
4
than it is in examples like thirteen men, Tennessee air, achromatic lens. This
prediction is a correct one, we believe. But examples like pretonic lengthening, goodlooking lifeguard are quite different. Retraction is strongly preferred, and the unretracted cases seem to be subject to the same elongation of interstress interval that
4
afflicts e.g. achromatic lens. Yet our system presently treats Montana cowboy and
4
(113) a.
8
6
7
1 23 4 5
Montanncowboy
w s w s
w
R
322
MARK
8
7
4
b.
6
2
LIBERMAN
AND
ALAN
PRINCE
good-lookinglifeguard
w sV
\s
w
R
The difficulty seems to be that the initial monosyllabicword (or # prefix) in cases
like good-looking is not being given its proper "weight". It seems wrong to give a
lexical entry (albeit monosyllabic)no greater representationin the metricalgrid than a
pretonic initial syllable would receive.
The difference in applicability of the Rhythm Rule is related to this greater
metrical strengthof the lexical monosyllable. If the grid column standingover good, in
(113), is made one level higher, and the scansion is adjusted in conformity with the
RPPR, then a stress clash will result, and the pressure for Iambic Reversal to occur
will be the same as it is in examples like thirteenmen:
11
(114)
*10
*9
6
7
8
1
2 3 4 5
good-l7oking lifeguard
X\ s w
\
w
R
It is reasonable that lexical entries (aside from unstressed function words) should
be guaranteed a certain metrical strength, even if they are monosyllabic, and the
addition of such a provision to our system results in an empiricallybetter account of
the Rhythm Rule's pattern of occurrence.
There are any number of formal devices one could employ to implement such a
provision. We propose to adopt the most straightforwardone, an addendumto the
Relative Prominence ProjectionRule stating that the main stress (if it exists) of any #level unit must have at least two levels of metrical strength. Like any other rule, this
one has effects outside the class of cases that motivate it. Designed to distinguishe.g.
ON
STRESS
AND
LINGUISTIC
RHYTHM
pretonic from Montana, it also distinguishesgym mat from gymnast, tentacle tip from
catamaran. The value of such distinctionsremainsto be explored; we are not aware of
Although we have argued that "relative prominence" is best treated as a constituentstructure relation, it is equally clear that syllables (or some other class of terminal
elements) differ among themselves in degree of stress. In the present treatment, the
inequalityamong stress numbers(whetherassigned by cyclic rules, or determinedfrom
relative prominence patterns by an algorithm like (12)) is replaced by the notion of
relative metrical strength, defined in terms of the metrical grid with which a relative
prominence pattern is aligned. The partial ordering of relative syllable stress thus
obtained is different, in certain respects, from that implied by cyclic reassignmentof
[1 stress] with concomitantstress subordination,or (equivalently)by the output of (12).
These differences, for the most part, represent an improvement.
Of course, a metrical grid can certainly be collapsed into a linear stress-number
representation,as we observed earlier. Therefore, preferencefor the metricaltheory's
picture of syllabic stress relationsdoes not help motivate the hierarchicalaspect of the
metricalgrid. However, a metricaltheory, like any other candidate in this area, must
provide a reasonable picture of relative syllable stress, and we believe that the gridalignment system just described, which was devised primarily to account for the
Rhythm Rule's pattern of application,passes this test ratherwell.
We will consider four cases: left-branchingstructureswith main stress on the left;
right-branchingstructureswith main stress on the right; left-branchingstructureswith
main stress on the right;and three-wordstructures, either right-or left-branching,with
medial main stress.
In a left-branchingstructurein which each constituenthas the relative prominence
pattern [sw], the RPPR requires only that the initial element be metrically stronger
than each of those that follow:
(115) 1
2 3 4 5 6 ...
X x
xX...
s
. . .
's\
MARK
324
LIBERMAN
AND
ALAN
PRINCE
Of course, (115) is merely the simplest grid alignmentthat meets the RPPR's wellformedness condition. There is nothing to prevent further differentiationamong the
sequence of "weak" elements, but neither is there anythingto requireit.
This pattern can occur both within the word and at higher levels:
(116) a. 4
1 23
execute
I I I
s ww
Vl
M
b.
5
1 2 3 4
knowledgeable
s w w w
sV
J
M
c. 13
9
10
11
12
1 2 3 4 5 6
7 8
law degree requirementchanges
s w
s
wS w s w
sw
,w
R
Some people feel that the last element in the sequence may be strongerthan the
other weak elements, a feeling that often surfaces, for example, in the way that such
patternsare set to music. However, the 1 4 3 2 patternthat the traditionalcyclic theory
assigns to cases like (116c) seems overly differentiated;the non-mainstresses are more
equal than this patternimplies. In cases like (116b), previous theories of course do not
differentiateamong the unstressed syllables any more than we propose to do.
ON
STRESS
AND
LINGUISTIC
RHYTHM
In cases like (116a), it may seem odd not to make the terminal syllable, which is
[+stress], metrically strongerthan the preceding unstressed syllable. However, Kiparsky (1977) argues that poets have treated such words exactly as our scansion suggests.
A right-branchingstructurein which each constituent has the relative prominence
pattern [ws] is the mirrorimage of the situationjust described, and our system treats it
in an analogous manner:
(117)
1
...6
5 4 3 2
... XXX
XX
. W www
W
vsgU
This pattern does not occur within the word, except as the result of successive
affixation of #-prefixes (which is a relatively rare occurrence); it is common enough,
however, in phrasal collocations:
VVV
(118) a.
10
7
8
9
12 3 4
56
chloride
poly vinyl
sw
w
s w
sw
ws
ply
vnlIIi cloridI
I
R
b.
1
5
4
XsXv~~~
R
326
MARK
LIBERMAN
c.
AND
ALAN
PRINCE
12
9
11
10
4 5 6 78
12 3
reports of threats of vi olence
w s
s ww
sW
S
s
Here again, the RPPR requiresonly that the strongest stress be metricallysuperior
to all the others. There is, however, a tendency to strengthen one of the otherwise
metrically equal positions. Thus most people pronouncepolyvinylchloridewith vinyl
less stressed thanpoly; it is interestingthat they also treat this word as if its constituent
structure were [[polyvinyl]chloride],which would produce such a pattern of relative
stresses as a matter of course:
(1 19)
12
10
11
7
8
9
12 34
56
poly vinyl chloride
sw
VVs
sw'
sw
Vs
\V/
w</
R
ON
STRESS
AND
LINGUISTIC
RHYTHM
(120)
7
5
1
2 3
4
John's three red shirts
w
w w
R
The "extra" grid element, labelled5 in the diagramabove, is perfectly consistent
with the RPPR, althoughnot requiredby it. It has the effect of breakingup a sequence
of otherwise equal "upbeats". The break-up of equal stresses into some kind of
alternating pattern seems very natural in the case of constituent-initialsequences,
"upbeats" so to speak, but quite unlikely in the case of posttonic sequences. Our
theory contains no left-right asymmetry,and thereforedoes not predict this distinction.
One possible alternativeis to define a kind of partialconstituent structureon metrical
grids, each unit extending from a given element to the next hierarchically equal
element. These units would be in a sense analogous to the metrical "feet" of writers
like Abercrombie,althoughthe particularsof this proposalare ratherdifferentfrom his.
According to this principle, (121a) would consist of only one "foot", while (121b)
would consist of three and (121c) of two:
(121) a. 4
1 2 3
x x x
w
b.
4
1 2 3
x x x
~~~~~w
w ,x
c.
6
4
5
1 2 3
In a case like John's three red shirts, making three metrically stronger than John's
and red would then cause the new "foot boundary" to fall on the major constituent
break, as well as breakingup the sequence of upbeats. If "foot"-like metricalunits can
be shown to be a phonetically useful concept, then this line of inquiry would become
quite appealing.
Note that some of the cases for which so-called "readjustmentrules" have been
devised have the right-branching[ws] patternjust discussed; e.g. (122):
328
MARK
(122)
LIBERMAN
AND
ALAN
PRINCE
14
11
12
13
5 6 7 8
1 2 3 4
9 10
thq cat that ate thq rat that stole the cheese
~w w
w s x1
w 5
w~~~~
5
R
Operating directly on the syntactically-motivatedsurface structure, our system
provides a counterpartto the even prenuclearstress patternthat the readjustmentrule
is designed to produce.19
The case of a left-branchingstructure in which each constituent has the relative
prominence pattern [ws] arises only at the phrasal level. Strings of possessives are a
common example:
17
(123)
*15
*16
*12
*13
*14
8
10
9
11
5 6
1 2 3 4
7
Sammy's father's brother'sdog
s
\/
w
s
/
w
s
I
I
w
R
19 Note, incidentally, that the normal phrasing this is the cat/that ate the ratlthat stole the cheese is
expected, independent of any constituent-structure readjustment, due to the fact that S boundaries in general
require intonation breaks.
ON
STRESS
AND
LINGUISTIC
RHYTHM
(124)
8
1
12
9
10
14
13
11
3 4
5 6
s
I
w
R
American history teacher, while in the meaning 'American teacher of history', parsed
2
MARK
330
10
13
11
12
234
5 67
(125) a.
1
LIBERMAN
AND
ALAN
PRINCE
Americanhistory teacher
W
S WW S
S WW
\sXr
svw S
Y\~~\W
w
s
b.
1
10
13
11
12
234
5 67
Americanhistory teacher
W
S WW
ws w
S WW
swws
sV sV
This article is based principallyon ideas about the representationof stress patterns set
out in Liberman (1975), a work that attempted primarily to construct a theory of
English intonational patterns, and to predict the effect of stress and constituent
structureon the alignmentof such tonal sequences with linguisticmaterial. Ourdebt to
other descriptions of English stress should be obvious; in particular,the phonological
system of SPE, and subsequentworks in the same tradition,forms the backgroundand
foundationof our research.
It has been brought to our attention that some aspects of our proposal are
prefiguredin Rischel (1964; 1972)and Martin(1972).
ON
STRESS
AND
LINGUISTIC
RHYTHM
332
MARK
LIBERMAN
AND
ALAN
PRINCE
primary-stressedsyllables of the string. But unless the examples are chosen with care,
considerable violence will be done to the true pattern of secondary stresses; for
20 Whichis essentially the constituentstructureimpliedby the concept of metricalsubdivisionin music,
as Martinpoints out.
ON
STRESS
AND
LINGUISTIC
RHYTHM
2
example, upon applicationof what we take to be his rules, the big dog emerges as the
3
big dog. 21
Althoughwe are in sympathywith Martin'saims, his system seems simplisticas a
description of rhythmic patterningin general, and of limited empirical value in the
description or generationof English stress patterns.
4. Conclusion
We have argued in this article for a substantialrevision in the theory of phonological
representation. The phonetic-phonological level of linguistic description has been
commonly conceived to provide only for simple concatenation of its basic unitssegments and boundaries.We believe, however, that many of the essential characteristics of stress systems can only be insightfully described in a theory that recognizes
hierarchy as a mode of phonological organization. In particular, we propose that
relative prominence is defined between phonological constituents, rather than on
individualsegments. Prominence, so defined, is projectedonto syllables by associating
them with a "metrical grid", which can be thought of as a hierarchy of intersecting
periodicities (ratherthan constituents),the structureof which is relevant to phenomena
of rhythmand timing.
Construingrelative prominencein this way leads to a numberof ratherunexpected
results. In earlier theories, rules of prominence assignment like the NSR, CSR, and
Schane's Detail Rule had to be equipped with essential variables, so that they could
have the power to search through unlimited stretches of string in order to find a
1-stressedvowel. In the present theory, such rules are renderedas very local conditions
on the labelling of nodes that are strictly adjacent in metrical structure. The metrical
versions of the CSR and NSR furthermoreapply to their entire domains simultaneously, rather than pursuing a cyclic path through them, as the SPE-or "linear"versions must. A theory that has variables and cyclic application allows for a
tremendous range of complex rules and rule interactions, as syntax shows us. Phrase
phonology does not exhibit this complexity, and the present theory excludes it in
principle by allowing rules to be local and simultaneous rather than cyclic and longdistance. The possibility of this restrictionis a direct reflex of the mode of representation we have adopted.
Extending the relational idea from the phrase level, where the relevant constituency is independentlydeterminedby syntax, to the word and below, we arguedthat
21 It may be that Martin intends the alignment process to be guided by the linguistically
observed
pattern of secondary stresses, as well as by the location of the "syllable marked for primary accent." This
interpretation would deny the "rhythmic hierarchies" any role at all in the assignment of stress patterns,
however, which does not seem to be Martin's position.
334
MARK
LIBERMAN
AND
ALAN
PRINCE
stress rules like the ESR function to confer the prosodically significant structure on
rows of syllables. Because of this, disjunction is seen to be a natural and necessary
concomitant of the application of stress rules, rather than an arbitraryproperty of
certain notations. Furthermore,from this perspective of function, the iterativityof the
ESR becomes quite intelligible, given that every syllable must be put in its metrical
place.
The structures induced by the ESR, motivated primarily by considerations of
relative prominence assignment, also allow a considerablesimplificationin the rules of
vowel reduction. Because the notion "weak constituent" is equally applicableto preand poststress positions, we were able to eliminate the special rule of Poststress
Reduction merely by stating the prestress rule in its simplest form. We also found that
the metrical conditioning of the English Destressing Rule was of the most elementary
sort, easily subsumed under the general restrictionthat stressless vowels occupy only
weak positions. This interpretation of the relationship between metrical strengthweakness and [+stress] plays a multifariousrole in the present theory, constrainingthe
application of such diverse processes as prominence assignment (tree-buildingalgorithm (57), LCPR (96)), destressing (EDR (70)), and readjustment of prominence
(Iambic Reversal (111)).
The phonological constituent structure required to support a simple mode of
assigning prominence patterns also proved to be basic to expressing their mutations
under rhythmicinfluence. The rule of Iambic Reversal, local in environment,simultaneous in its application,describes the essential form that rhythmicstress shift assumes
in English. The factors that encourage the application of this optional rule are,
however, of a rather different kind from those at play in the initial construal of
relational patterns. The basic notion of a motivating "stress clash" is not one that is
perspicuously definable on constituent structure. To model it, and its desirable
opposite "alternationof stress", we formalizeda second kind of hierarchy,a stratified
grid with a notion of metricalstrengthintrinsicto it. When materialis associated to the
grid via the Relative ProminenceProjectionRule (104), which is based on the strongweak relation between constituents, stresses that are intuitively "clashing" appear
adjacent at the same grid level. Judicious application of Iambic Reversal eliminates
these adjacencies. Other factors that appear to influence Iambic Reversal, such as
interstress distance and the status of "upbeats", are directly represented in the grid
formalism.
Our proposal entails an enrichmentof phonologicaltheory. But it is strengthening,
ratherthan a dissipation of our understanding.By removingfrom linear representation
those aspects of stress and rhythm that are essentially hierarchical in nature, and
providingthem with a truly appropriateformal characterization,we lay the groundwork
for a greatly simplified and narrowly restricted theory of stress phenomena and
prosodic processes.
ON
STRESS
AND
LINGUISTIC
RHYTHM
References
Aronoff, M. H. (1976) Word Formation in Generative Grammar, MIT Press, Cambridge,
Massachusetts.
Austin, T. R. (1976) "A Non-cyclic Analysis of German Word Stress," Occasional Papers in
Linguistics II, University of Massachusetts at Amherst, 105-116.
Bing, J. (1976) "Is the Cycle Necessary in Phonology," unpublished paper, University of
Massachusetts, Amherst, Massachusetts.
Brame, M. K. (1972) "The Segmental Cycle," in M. K. Brame, ed., Contributions to Generative
Phonology, University of Texas Press, Austin and London, 62-72.
Bresnan, J. (1972) "Stress and Syntax: A Reply," Language 48, 326-342.
Chomsky, N. (1951) Morphophonemics of Modern Hebrew, unpublished Master's thesis,
University of Pennsylvania, Philadelphia, Pennsylvania.
Chomsky, N. and M. Halle (1968) The Sound Pattern of English, Harper and Row, New York.
Fischer-J0rgensen, E. (1948) "Some Remarks on the Function of Stress with Special Reference
to the Germanic Languages," Congr. Intern. Sc. Anthropol. and Ethnol., ComptesRendus, Ille Session, Bruxelles 1948, 86-88.
Gimson, A. C. (1962) An Introduction to the Pronunciation of English, St. Martin's Press, New
York.
Goldsmith, J. (1976) Autosegmental Phonology, unpublished Doctoral dissertation, MIT,
Cambridge, Massachusetts.
Jones, D. (1956) An Outline of English Phonetics, W. Heffer and Sons, Cambridge.
Kenyon, J. S. and T. A. Knott (1953) A Pronouncing Dictionary of American English, G. and C.
Merriam, Springfield, Massachusetts.
Halle, M. (1973) "Stress Rules in English: A New Version," Linguistic Inquiry 4, 451-464.
Halle, M. and S. J. Keyser (1971) English Stress: Its Form, its Growth, and its Role in Verse,
Harper and Row, New York.
Kean, M. L. (1974) "The Strict Cycle in Phonology," Linguistic Inquiry 5, 179-203.
Kahn, D. (1976) Syllable-Based Generalizations in English Phonology, unpublished Doctoral
dissertation, MIT, Cambridge, Massachusetts.
Kiparsky, P. (1966) "Ueber den Deutschen Akzent," Studia Grammatica 7, Akademie-Verlag,
Berlin.
Kiparsky, P. (1973) "Elsewhere in Phonology, " in S. R. Anderson and P. Kiparsky, eds., A
Festschrift for Morris Halle, Holt, Rinehart, and Winston, New York.
Kiparsky, P. (1977) "The Rhythmic Structure of English Verse," Linguistic Inquiry 8, 189-247.
Liberman, M. (1975) The Intonational System of English, unpublished Doctoral dissertation,
MIT, Cambridge, Massachusetts.
Liberman, M. and L. Streeter (1976) "Use of Nonsense Syllable Mimicry in the Study of
Prosodic Phenomena," paper delivered at Acoustical Society of America meeting, San
Diego, Nov. 1976.
Marchand, H. (1969) The Categories and Types of Present-Day English Word-Formation, C. H.
Beck'sche Verlagsbuchhandlung, Munchen.
Martin, J. (1972) "Rhythmic (Hierarchical) versus Serial Structure in Speech and Other
Behavior," Psych. Review 79, 487-509.
Mascaro, J. (1976) Catalan Phonology and the Phonological Cycle, unpublished Doctoral
dissertation, MIT, Cambridge, Massachusetts.
Nakatani, L. and J. Schaffer (1976) "Hearing Words Without Words: Prosodic Cues for Word
336
MARK
LIBERMAN
AND
ALAN
PRINCE
Perception," paper delivered at Acoustical Society of America meeting, San Diego, Nov.
1976.
Nanni, D. L. (1976) "Vowel Destressing in English," unpublished manuscript, University of
Massachusetts, Amherst, Massachusetts.
Oehrle, R. T. (1971) "Some Remarks on the Role of Morphology in the Assignment of Stress,"
mimeographed paper, MIT, Cambridge, Massachusetts.
Prince, A. (1975) The Phonology and Morphology of Tiberian Hebrew, unpublished Doctoral
dissertation, MIT, Cambridge, Massachusetts.
Rischel, J. (1964) "Stress, Juncture and Syllabification in Phonemic Description," Proceedings
of IXth International Congress of Linguists, 1962, 85-93.
Rischel, J. (1972) "Compound Stress in Danish without a Cycle," Annual Report of the Institute
of Phonetics, University of Copenhagen (ARIPUC) 6, 211-228.
Ross, J. R. (1972) "A Reanalysis of English Word Stress," in M. K. Brame, ed., Contributions
to Generative Phonology, University of Texas Press, Austin and London, 229-323.
Schane, S. (1972) "Noncyclic English Word Stress," mimeographed paper, University of
California at San Diego, La Jolla, California.
Siegel, D. (1974) Topics in English Morphology, unpublished Doctoral dissertation, MIT,
Cambridge, Massachusetts.
Steele, J. (1975) An Essay towards Establishing the Melody and Measure of Speech, facsimile
edition, Scolar Press (1969), Menston, England.
Trager, G. L. and H. L. Smith (1951) An Outline of English Structure (Studies in Linguistics,
Occasional Papers 3), Battenburg Press, Norman, Oklahoma.
Vergnaud, J.-R. (1974) Problemes Formels en Phonologie Generative, These de 3e Cycle,
Universit6 de Paris VII, Paris, France.
Vergnaud, J.-R. (1976) "Formal Properties of Phonological Rules," unpublished paper,
C.N.R.S., Paris, France.
Zonneveld, W. (1976) "Destressing in Halle's English Stress Rules," Linguistic Inquiry 7, 520525.
Liberman
Bell Telephone Laboratories
2C-302A
Murray Hill, New Jersey 07974
Prince
Department of Linguistics
South College
University of Massachusetts
Amherst, Massachusetts 01003